A Signaling Theory of Distributive Policy Choice: Evidence From Senegal∗ Jessica Gottlieb
†
Guy Grossman
‡
Benjamin Marx
Horacio A. Larreguy
§
¶
November 12, 2016
Abstract A recent literature emphasizes political economy factors behind the wave of administrative splits across the developing world. While past studies have focused on why some groups are more likely to obtain new administrative units, they do not explain why vote-maximizing incumbents use this arguably wasteful policy in the first place. We contribute to this literature by embedding administrative unit splits within incumbents’ broader electoral strategy of distributive policies. We argue that incumbents prefer to target local public goods to groups for whom this is a credible signal of commitment, namely those with a history of a reciprocal relationship. Other groups require a stronger signal which, we argue, is emitted by the creation of new administrative units that entail an increase in stable fiscal transfers due to the stickiness of administrative boundaries. We test our signaling theory using the case of Senegal, and find robust support for its core predictions.
∗ We
thank participants at the 2016 NOVAFRICA Conference on Economic Development in Africa for helpful comments. We are grateful to the National Statistics Agency, the Independent Electoral Commission, and the Direction of Local Governments in Senegal for generously sharing their data. We are indebted to Assane Ba and Nina Marx for their invaluable help during the fieldwork, and we thank Shelley Liu and Miguel Eusse for outstanding research assistance. † The Bush School of Government & Public Service, Texas A&M University.
[email protected]. ‡ Department of Political Science, University of Pennsylvania.
[email protected] § Department of Government, Harvard University.
[email protected]. ¶ Department of Economics, Massachusetts Institute of Technology.
[email protected].
1
Motivation
In the past two decades, many developing countries have responded to pressures to decentralize administrative authority by increasing the number of subnational units (Grossman et al., n.d.). The ubiquity of this dramatic reorganization of the territorial structure of states has led to a growing body of work on the determinants of such administrative unit splits. Specifically, the current literature has rejected functionalist explanations of the creation of administrative units—those rooted in efficiency tradeoffs—in favor of political economy explanations, which are generally based on the electoral benefits that splits confer on national incumbents (Pierskalla, 2016b). We advance this newer line of reasoning by addressing two remaining puzzles in the literature: 1) when and why incumbents choose administrative unit provision over a menu of other distributive policies, and 2) why incumbents risk the electoral benefit from voters in newly created units when they may face punishment from voters and local politicians in rump units. Previous work argues that national incumbents—especially those facing heightened electoral pressure—pursue such reforms to: create local public sector employment that both co-opts local elites in newly created administrative units and provides patronage to lowlevel party functionaries (Green, 2010; Hassan, 2016); divide the power of the opposition (Malesky, 2009; Resnick, 2014); reduce the bargaining power of the periphery vis-àvis the center (Grossman and Lewis, 2014); and increase the executive branch’s control of parliament and the surveillance of the electorate (Hassan and Sheely, 2017). However, while the overall effect of dramatically increasing the number of administrative units on public service delivery is ambiguous at best (Pierskalla, 2016b), it undoubtedly entails significant pecuniary and non-pecuniary costs for incumbents, and risks alienating some voters who oppose such splits. This raises a natural puzzle: why would incumbents facing electoral pressure engage in an arguably costly and wasteful policy, designed to achieve narrow electoral (partisan) goals? Past studies attempt to address this puzzle by arguing that voters from newly created units favor such a policy. This is because splits reduce voters’ 1
distance from the administrative unit’s headquarters, and increases local control over central government transfers (Grossman and Lewis, 2014), or because targeted patronage steers the local economy (Hassan, 2016). Incumbents use this policy to especially target marginalized groups (Kimura, 2012) that place a relatively high premium on new administrative units, due to a strong preference for self-governance in their homelands (Hassan, 2016) or because the status quo contributes to their marginalization (Grossman and Lewis, 2014). While past work has undoubtedly increased our understanding of the dynamics of administrative unit splits, we argue that the above attempts to solve the core puzzle of administrative unit splits is wanting. First, and most important, past studies, have not addressed an auxiliary core puzzle: if incumbents want to lure certain voters, and if creating many new administrative units is costly (and arguably wasteful), why then don’t incumbents simply use an alternative strategy—local public goods—that is both valued by voters and contributes unambiguously to development? In other words, past work has generally ignored the fact that incumbents have a menu of electoral strategies to lure voters, and it is unclear why they would choose to increase the number of administrative units, rather than to adopt a different targeting strategy. Second, even if voters from newly created administrative units place a high premium on local benefits of newly created units and reward the incumbent in subsequent elections, the preferences of voters and local opinion leaders from rump areas—potentially opposing splits—should also affect incumbents’ policy decisions. However, the current literature has close to nothing to say about such preferences, even though incumbents facing electoral uncertainty must surely consider their perspective when splitting an administrative unit.1 This knowledge gap is problematic, since the preferences of all voters (not just those in newly created units) should determine whether incumbents want to pursue those reforms. We address these gaps by proposing a new theoretical framework for understanding both 1 Both
Grossman and Lewis (2014) and Hassan (2016), report that incumbents did not suffer a drop in mean support in rump communities in the elections that followed district splits in Uganda and Kenya, respectively. However, neither study offers a rigorous account of rump areas that theorizes about voter preferences in these areas or disaggregates them spatially as we do.
2
the ubiquity and differential use of administrative unit splits in the past two decades. Our starting point is that in low-information electoral settings, voters search for signals (heuristics) to determine politicians’ congruence. In settings characterized by weakly institutionalized, non-ideological political parties, congruence is commonly defined by the extent to which candidates will take the interest of constituents to heart if they assume or continue in office (Chandra, 2004). We further assume that in such settings, voters do not have strong attachments to (non-programmatic) political parties, and thus can switch with relative ease to the party or candidate that sends the clearest signal (Key, 1966; Gottlieb and Larreguy, 2016). Incumbents understand the importance of such signals, and are well-positioned to use distributive policies strategically. Specifically, incumbents have two core policies to signal to voters (and perhaps more importantly, to their brokers) a strong commitment to their welfare: (a) local public goods, and (b) new administrative units.2 We argue that incumbents prefer using local public goods, which are less costly and generally more effective, but face a core problem of establishing credibility with some, but not all brokers or the voters they coordinate. We consider two group-level factors affecting the incumbent’s choice of pre-election distributive policies: (a) the coordination capacity of groups’ brokers, and (b) the history of a reciprocal relationship between a group and the incumbent party. Within this framework, we formulate several hypotheses pertaining to the incumbents’ arbitrage between providing local public goods and creating new administrative units. First, a vote-maximizing incumbent has an incentive to only target areas where brokers are strong enough to effectively coordinate votes around a single candidate. Second, when targeting groups with strong brokers, incumbents prefer making (promises of) public goods , but only when their party has a history of reciprocal relationship with those groups. We argue that with such groups, public goods emit a signal strong enough to sustain an exchange equilibrium, whereby the group 2 Admittedly,
incumbents can employ additional strategies such power-sharing (Francois et al., 2015), elites’ appointments to ministerial positions (Arriola, 2013), or passing laws that have a strong regional and/or ethnic appeal (Lieberman, 2003). These alternative strategies operate at the national level, while in this project our interest is in spatial variation in targeting at a more local level.
3
maintains its electoral support for the incumbent, who, in turn, continues his targeting to the group in the post-election period. Finally, when strong broker groups do not have a history of a reciprocal relationship with the incumbent, they may not view pre-election local public goods (or promises of such goods after the election) as a credible signal of post-election commitment. The incumbent thus targets such groups with new administrative units. Key to our argument is the notion that, unlike public goods flows that can be phased in or out with relatively ease, the creation of an administrative unit entails both short-term upfront costs of setting up a new local administration, as well as an increase in stable fiscal transfers due to the relative stickiness of administrative boundaries and fixed unit-level outlays. This variability in the incumbent’s ability to send a credible pre-election signal of congruence leads to clear testable implications. First, incumbents should use different strategies vis-à-vis different types of groups. Second, we expect members of groups targeted with administrative units to reciprocate by increasing their vote share for the incumbent, compared to groups that are targeted by the incumbent neither administrative units nor transfers.3 We test these predictions using fine-grained original data from Senegal, an ideal context in which to study the strategic targeting of administrative splits and local public goods, as well as citizens’ responses to incumbents’ differential targeting policies. First, Senegal is a young democracy exhibiting multiparty competition,4 ample party switching and group-level targeting (Koter, 2013). Second, Senegal recently went through a series of dramatic administrative unit changes. Following his re-election to a second term in 2007, president Wade of the Parti Démocratique Sénégalais (PDS) split a large number of low-level local government units, communautès rurales (or rural communities, henceforth CRs). By splitting existing CR units, Wade’s government created 74 new ones in 2008 and 16 in 2010-2011, which affected 18% of the country’s villages, as reflected by Figure 1 and Appendix Table A.5. We find robust evidence in support of our theoretical predictions when applied to our 3 In
contrast, as explained later, it is unclear whether the targeting of public goods should lead increased electoral support for the incumbent by groups that were already supporting it. 4 The Parti Démocratique Sénégalais (PDS) ended the long-ruling of the Socialist Party (PS) in 2000 to lose power to the Alliance for the Republic (APR) in 2012.
4
Figure 1: CR Changes in 2008
Note: “Not in sample"” correspond to Dakar and St Louis urban areas.
case, and argue that similar findings should obtain in other low-income electoral regimes. First, we show that the incumbent party (Wade’s PDS) is using different distributive policies to target different groups. On the one hand, Wade’s administration is significantly more likely to target public goods to villages belonging to a group with strong brokers and a strong track record of a reciprocal electoral relationship – the Mouride religious brotherhood (O’Brien, 1975; Boone, 2003b). On the other hand, President Wade is significantly more likely to grant new administrative units to an ethno-linguistic group with strong brokers but a weaker history of ties to the incumbent party: the Toucouleur (Beck, 2008). Second, using a difference-in-difference estimation strategy, we demonstrate that administrative unit creation is an effective targeting strategy, driving a large increase in support for the PDS in the election immediately following the creation of new CRs both in new CRs but also in villages in rump CRs located far away from the headquarters. This paper makes several important contributions to the nascent, yet growing, literature on administrative unit splits in the developing world. Most importantly, we embed administrative splits within a larger political economy framework of distributive politics. Past 5
studies have all advanced theoretical explanations of administrative unit splits that treat the policy in isolation from other policy instruments. This is problematic not simply because targeting policies are likely substitutes, but also because previous accounts cannot explain why incumbents adopt a very costly and allegedly wasteful distributive policy to begin with. Our paper contributes to the literature by proposing a more general theory of policy choice. Specifically, we advance the literature by offering a novel argument that links administrative unit creation to the credibility (or signal strength) of an incumbent’s congruence or longterm commitment to public service delivery. While others have argued that administrative units are targeted to certain groups for reasons particular to the country context, our theory is more general, giving it explanatory power outside our chosen case and in contexts of clientelistic democracies where groups of voters vary in both their ability to coordinate votes and their demonstrated reciprocal relationship with the incumbent party. This paper also contributes to the literature on the effect of growing electoral pressure in Sub-Saharan Africa on the types of policies that incumbents adopt; such policies are generally visible, salient, popular and easy to implement, and can relatively easily be attributed to the incumbent (Grossman, 2015; Harding and Stasavage, 2014). Finally, it contributes to a body of work on the long-term effect of the administrative structures that were put in place by the colonizers of Africa (Englebert, 2000). The rest of this paper is organized as follows. We present our theoretical argument in Section 2, and provide relevant background on our empirical case in Section 3, focusing specifically on the variation in broker strength and reciprocal ties to the incumbent among distinct groups in Senegal. In Section 4, we describe the estimation framework we use to test our core hypotheses on the effect of splits on per capita transfers, targeting, and the electoral consequences of splits. Section 5 presents our main empirical results, and Section 6 concludes.
6
2
Theoretical Argument
In this section we develop our theoretical argument. We start by laying out several core assumptions about the nature of political competition and distributive politics in many lowincome countries. We embed our discussion in the context of African politics, though contend that our theoretical argument is likely relevant to other regions, especially those with poor information access where politicians need to signal congruence through targeted transfers and the availability of brokers that help coordinate groups of voters. Though falling short of expectations, there is ample evidence suggesting that the introduction of multi-party elections in the early-1990s across Africa, and elsewhere, has incentivized national incumbents to adopt policies with a relatively wide appeal (Harding and Stasavage, 2014). Political competition in many developing countries, however, exists alongside parties that are weakly institutionalized and, for the most part, non-programmatic (Riedl, 2014; Lupu, 2016). Importantly, contrary to some simplistic depictions of elections in Africa as “ethnic censuses,” a large share of voters are “uncommitted,” or non-partisan.5 Furthermore, there are good reasons to reject the idea that voters care only, or even mostly, about petty clientelistic transfers (Fujiwara and Wantchekon, 2013; Weitz-Shapiro, 2014). Instead, we assume that the majority of voters care about incumbent’s congruence, as defined above. Given the non-programmatic nature of politics, this is manifested in the commitment of incumbents to making targeted transfers to particular groups of voters, such as public services (Carlson, 2015).6 Elections in Africa, however, take place in a low-information environment.7 This has important implications for the strategies of both voters and politicians. Voters have an 5 Data
on the extent to which voters are non-partisan comes from voter reasoning surveys (Weghorst and Lindberg, 2013) and from actual election data at the polling station level (Gottlieb and Larreguy, 2016). 6 While this does not fit individualized forms of clientelistic exchange (e.g., Stokes (2005)), it is related to Kitschelt and Wilkinson’s (2007) idea of “collective clientelism” in which parties target collective transfers to groups in exchange for electoral support. See also the work of Gingerich and Medina (2013) and Rueda (2016) in Brazil and Colombia, respectively. 7 Media outlets in developing countries often lack capacity to adequately report about politicians’ performance and credentials (Moehler and Singh, 2009), and/or are prone to misrepresenting politicians’ actions due to capture or partisan bias (Boas and Hidalgo, 2011).
7
incentive to follow cues from local opinion leaders (or brokers) when politicians reward bloc voting villages. This is especially the case if brokers have superior information regarding candidates’ ‘type’ (Baldwin, 2013) and a high capacity to coordinate voters around one candidate (Rueda, 2016). Candidates therefore have a strong incentive to signal to local brokers that, if elected, they would take the interests of their communities to heart. Since pre-election promises are not always credible, candidates may need to signal congruence with voters using actual transfers. Controlling the national coffers, incumbents have an advantage over challengers, as they can adopt distributive policies to signal congruence to brokers and voters (Collier and Vicente, 2012). A core assumption of this study’s theoretical argument is that different distributive policies emit signals with different strengths, and that incumbents choose policies strategically, depending on the strength of the signal that is needed to lure brokers and their voters.
2.1
Incumbents’ targeting strategy
We consider a society that is made up of three groups. Group A has weak brokers, in the sense that they have a hard time coordinating voters around candidates. Naturally, incumbents have a weak incentive to target such group (Gottlieb and Larreguy, 2016).8 By contrast, both groups B and C have brokers with a high capacity to coordinate voters around candidates who send a strong enough signal of commitment. Naturally, vote-maximizing incumbents have a strong incentive to target such groups. Groups B and C, however, differ on one key dimension:
a history of a reciprocal rela-
tionship with the incumbent. By this we mean that a group has received targeted transfers from, and previously voted in support of, the party of the incumbent. In short, we argue that following through on past promises makes promises of future transfers more credible. For such groups, targeted transfers in pre-election years are more likely to suffice as a cred8 Notably,
even in high information environments, voters have a hard time interpreting signals from the level of service provision in their locality (Clinton and Grissom, 2015). Local opinion leaders thus play an important mediating function, in both high and low-information environments.
8
ible signal of congruence. In Section 3, we use the Senegalese case to illustrate potential determinants of such a reciprocal relationship, including shared identity with the incumbent and economic autonomy from the state. By contrast, Group C does not have a history of a reciprocal relationship with the incumbent, and thus its brokers may not view pre-election public goods as a credible signal for post-election commitment. This is because such flows can be phased in or out with relative ease. Thus a stronger signal is needed for luring the brokers of Group C to coordinate the individuals under their influence to vote for the incumbent. We argue that incumbents grant groups of voters in Group C new administrative units to send a strong signal of congruence; i.e., of post-election commitment. Implicit in our discussion is the assumption—for which we provide evidence for in Section 5—that administrative unit splits are rather costly: they increase the size of the local bureaucracy and the flow of stable per capita transfers. We formalize the above discussion of the incumbent’s targeting strategy in the following hypotheses: H1a Strong-brokers targeting: incumbents disproportionally target groups with “strong” brokers, compared to groups with “weak” brokers. H1b Strong-broker group with history of reciprocal relations: among groups whose brokers have a relatively high coordinating capacity, incumbents disproportionally target public goods to groups with whom they have a history of mutual reciprocal exchange. H1c Strong-broker group with little or no history of reciprocal exchange: among groups whose brokers have a relatively high coordinating capacity, incumbents disproportionally target administrative unit splits to groups with whom they do not have a history of mutual reciprocal exchange.
9
2.2
Voters’ preference for administrative unit splits
Above we argue that granting a group a new administrative unit sends a strong signal of post-election commitment. Why might this be the case? Building on the existing literature on administrative units proliferation, and on qualitative fieldwork we conducted in Senegal,9 we briefly discuss some of the reasons voters express a strong preference for being granted their own administrative unit (via splits).10 First, voters’ preference for carving out their own administrative unit has generally increased following decentralization reforms, since the transfer of responsibilities and resources to subnational tiers of government make the control of such units ever more consequential (Grossman and Lewis, 2014). Especially where local governments are financed almost exclusively by central government transfers, being granted a new local government entails a significant increase in fixed fiscal transfers. Grossman et al. (n.d.) demonstrate that such transfers have important redistributive consequences: when they increase the level of local public services, this generally comes at the expense of rump areas. Key to our argument is the idea that an increased stream of central government transfers to new local governments is stable over time due to the relative stickiness of administrative boundaries compared to other policy instruments. Second, in developing countries, which typically have low-capacity local governments, voters further support splits because these unambiguously increase administrative attention. Administrative attention captures local governments’ limited ability to service a large number of residents (especially given the in-person nature of the interaction between citizens and public officials), as well as the difficulty in servicing far-flung villages (e.g., monitoring, training and stocking frontline public service points). Splits increase administrative attention by reducing the number of residents and villages that need to be serviced, and the average distance between a local government’s headquarters and the areas it serves. In other words, 9 See
the Appendix for further qualitative evidence in support of our assumptions about voter preferences. preference for local public services is self-evident, and well documented in the literature.
10 Voters’
10
administrative unit splits limit how far a local civil service has to stretch its limited resources and bureaucratic reach to outlying (peripheral) villages. Since low-capacity governments can pay more attention to villages located close to their headquarters, more distant areas benefit most from administrative unit splits. This is because, for these areas, splits decrease villagers’ traveling distance to the local government headquarters (Grossman and Lewis, 2014), and help steer the local economy (Hassan, 2016). Indeed, for both political and practical reasons (most power brokers and public services are located in or near the local government’s headquarters), the value of a new administrative unit increases the further one resides from the old local government’s headquarters. We note that administrative unit splits come at a cost. Against the benefits of greater central government transfers per unit and of larger administrative attention, citizens in more remote areas must weigh the possibility that splits will result in a new local government that is less economically viable, that is less able to take advantage of economies of scale, and that may have a hard time filling-up vacancies, at least in the short to medium term (Lewis, 2014). Villages that are located far from the local government headquarters and vie for a new administrative unit are nonetheless likely to support a split because they expect to benefit the most (and risk losing the least) from greater administrative attention. We formalize the above discussion with the following hypothesis: H2 New Local Government: areas that receive a new local government (via splits) reciprocate by voting for the incumbent in the next election. Naturally, residents in “rump” areas, (i.e., the part of an original administrative unit that remains after a new one is created), must also weigh the pros and cons of splits. In the aggregate, voter support should be lower than in remote areas that receive a new administrative unit. Leaders in rump areas may also oppose splits because they lose control over a large share of the territory of their constituency and may be subject to earlier re-election. We argue, however, that the preference for splits in rump areas is strongest where the expected
11
benefits from increased attention are largest—i.e., villages located furthest from the old administrative unit’s headquarters. The above discussion leads to an additional hypothesis: H3 Rump areas and Distance to headquarters: voters in rump areas perceive fewer benefits from splits compared to those in areas receiving a new local government and thus vote for the incumbent at lower rates. In such areas, the electoral reward to the incumbent post-split increases with distance to the local government’s headquarters. In addition to signaling a commitment to voters, administrative unit splits could also increase the surveillance of (and ability to mobilize) the electorate. In both cases, we would expect splits to increase the incumbent’s vote share more than would be expected in response to immediate transfers of goods. In Section 5, we provide evidence that adjudicates between these alternative explanations.
3
Political and Social Context
In this section we provide the necessary background on political and administrative decentralization in Senegal to contextualize the distributive policy choice faced by the incumbent and apply our theory of differential targeting by group type to distinct ethno-religious groups in Senegal. Senegal offers an ideal context in which to study incumbents’ strategic use of different distributive policies and the effects of administrative unit creation on electoral outcomes. First, social and religious groups in Senegal have brokers with varying degrees of voter coordination capacity (Gottlieb, 2016) and reciprocal ties to the incumbent. Second, political competition in Senegal is relatively high; since alternation of power via the ballot box is possible, Senegalese incumbents must adopt distributive policies wisely or face an electoral defeat.11 Third, Senegal recently witnessed a series of widespread administrativeunit splits: about 20% of villages were affected by splits in 2008 alone; a pre-election year. 11 While
Senegal legalized multiparty competition in 1990, electoral competition was hampered by the ruling party’s control over the electoral process. In 2000, the first democratic transfer of power (from the PS to Wade’s PDS) took place.
12
Fourth, the president of Senegal has almost total control over splits, which allows us to better focus on targeting as opposed to analyzing splits that reflect voters’ choice. Fifth, Senegal makes available fine-grained data at a very disaggregated level—village or polling station—over time, which allows us to improve on identification strategies used in previous studies, as explained below.
Decentralization in Senegal: Historical and Legal Aspects With the exception of few major cities, Senegal did not have formal local governments until 1972, and did not elect local representatives with executive power until 1990. Since the 1990s, however, the pace of decentralization has increased dramatically. A 1992 law established regions as a new tier of government, and a 1996 reform transferred executive powers to regions and the three lower local government tiers: towns (communes), municipalities within the country’s five largest cities (communes d’arrondissement), and CRs.12,13 The 1996 law provided the regulatory framework for administrative unit splits during the period covered in this study. Though the creation of new CRs was subject to the advisory opinion of regional councils, it ultimately entered into force only through a government decree signed by the president or prime minister who were not liable to provide justifications for splits.14 And, while the law also stated that prior to changes in administrative boundaries, the opinion of “all interested rural councils, municipal councils, and regional councils [was] required,” it was not explicitly binding. In 2010, the minister in charge of decentralization stated that “the government can reserve the right to create a commune, a rural community, a region or a département wherever it deems necessary” (Le Soleil, October 2010).15 12 Most
of the literature has viewed these reforms as furthering the interests of the “PS state” because they strengthened local patronage networks. Boone (2003a), for example, argues that Senegal’s decentralization was part of an institution-building strategy of power sharing that allowed both the central government and local elites to extract more rents. Nevertheless, others (e.g., Dahou and Foucher (2009)) have argued that decentralization triggered a de facto dispersal of resources that made it easier for opposition parties to emerge—including Wade’s PDS, which in many areas was able to successfully capture the PS clientèle. 13 A more recent reform (2014) suppressed regional councils, transferred more powers to the départements, and harmonized the status of towns and CRs to create a single commune status. 14 Article 193 du Code des Collectivités Locales de 1996. 15 Following a split via government decree, rural councils were legally dissolved and the CR would be
13
Social Setting Existing narrative accounts describe two culturally distinct groups in Senegal—the Mouride religious brotherhood and the ethnic Toucouleur—as having notably influential local leaders that serve as vote brokers (Boone, 2003b; Beck, 2008; O’Brien, 1975). Gottlieb (2016) shows empirically that villages with higher concentrations of these groups are more likely than other villages to coordinate votes. While both groups have relatively strong brokers with a high capacity to coordinate votes, they are distinct in their prior history of a reciprocal relationship with Wade’s PDS. As we explain below, the Mouride resemble Group B with a history of a strong reciprocal relationship with the incumbent; the Toucouleur resemble Group C without such a history. Two factors serve as potential determinants of a reciprocal relationship with a political party or candidate: shared identity and economic autonomy. First, co-ethnicity and coreligiosity are frequently cited in the extant literature as drivers of reciprocal relationship with politicians either because shared identity serves as a heuristic for candidate quality in information-poor contexts (Conroy-Krutz, 2013) or because it triggers expectations of favoritism (Chandra, 2004). Second, economic autonomy can also support a reciprocal political relationship, though this is somewhat less intuitive. For electoral reciprocity to precede an incumbent’s rise to power or continue after he loses an election, voters and their brokers must be willing to be in the political opposition, at least temporarily. We argue that greater economic autonomy—i.e., less reliance on the government for economic well-being—can make a group or its brokers take a long view, and support opposition candidates or parties and thus sustain reciprocal relationships outside an incumbent’s reign. The Mouride—the second largest Sufi brotherhood in Senegal—are generally considered the most loyal partisans of Wade.16 This is, in part, due to Wade’s membership in the administered by a “special delegation” until local elections could be organized. The automatic removal of local elected officials regularly triggered conflicts. For example, the military had to be deployed to install the “special delegation” in Chérif Lo, and in Mbane councilors went on a hunger strike to express their opposition to a split (Sud Quotidien, May 24, 2011). 16 The Toucouleur belong almost entirely to the largest brotherhood, known as Tidjane, making these
14
brotherhood and to the public attention he lavished on the brotherhood’s influential leadership (Resnick, 2013). In addition, the Mouride’s strong political brokers have traditionally been the most economically autonomous from the state (Boone, 2003b; Beck, 2008).17 The Mouride thus have greater capacity than other groups to support an opposition candidate, which they did in the 1993 elections when many of their religious leaders supported Wade’s PDS (Beck, 2008). Further, after the fall of the PDS from presidential power in 2012, the 2014 local elections saw continued support for the PDS in both the Mouride holy city of Touba and the province of Mbacké in which it is located. Notably, the region that is home to these two places is the only one of 14 where Macky Sall’s 2016 referendum was voted down; as PDS leaders encouraged this “no” vote as a plebiscite on Sall’s presidency, the Mouride were again squarely in the opposition (Kelly, 2016). Turning to the Toucouleur, both Boone (2003b) and Beck (2008) explain broker strength, and thus vote coordinating capacity, among the Toucouleur as deriving from a hierarchical social structure enshrined in a caste system. In contrast to the Mouride, Beck (2008) identifies these brokers as “dependent” upon the incumbent regime because they have access to fewer resources.18 It can be argued that the Toucouleur have insufficient financial autonomy to form loyalties to any particular political party and must instead negotiate opportunistically for credible promises of transfers, generally from the incumbent. The resulting unwillingness to join the opposition is evidenced by the relatively high level of electoral support in 2000 for the outgoing incumbent party (PS) in the most densely Toucouleur province, Matam (71 percent), compared to the relatively high level of support for the opposition (PDS) in the most densely Mouride province, Mbacké (63 percent). categories nearly mutually exclusive. 17 O’Brien (1975) attributes the strength of Mouride leaders to their status as the dominant local authority structure following the collapse of the pre-colonial state. During and after colonization, Mouride religious leaders or marabouts were the main intermediaries between the peasants of Senegal’s populous groundnut basin and the state. 18 Boone (2003b) traces this dependency to an economic crisis in the 1970s brought on by a severe drought that shifted “socioeconomic power away from the old elite” (306), which then sought cooperation with later state-led development initiatives in order to re-entrench their own power.
15
4
Empirical Framework
In this section we describe our empirical strategy for three analyses: (a) the effect of CR splits on central government transfers; (b) the relationship between distributive policy targeting and group identity; and (c) the effect of CR splits on incumbents vote share. We elaborate on these analyses below, in turn.
4.1
New CR Creation Effects on Central Government Transfers
We begin by testing whether, consistent with voters’ expectation described in Section 2, administrative unit splits entail an increase in future financial flows for affected communities. To that end, we use data on transfers from the central government to CRs for the period 2007-2014, which we obtained from the Division for Local Governments (Direction des Collectivités Locales), of the Senegalese Ministry of the Interior. Our main measure of transfers aggregates two types of financial flows: (1) current expenditures, and (2) long-term investment projects.19 Using these data, we run the following fixed effects specification:
transf ersijt = α + βSplitjt + δt + εijt
(1)
where transf ersijt denotes per-capita transfers received by CR i contained in original CR j, in year t (measured in levels and in logs), Splitjt is a dummy for split CR (defined at the level of original CR j), and δt are year fixed effects. Standard errors are clustered at the j level. Note that we cannot include CR fixed effects ηi in this specification since transfers are not observed at the level of rump and new CRs prior to the split. We are further interested in testing whether transfers vary between rump and new units after a CR split. We use information culled from government decrees to code whether a village is part of a new (or rump CR). Villages are assigned to a rump CR if they continue under the administrative jurisdiction of the old CR headquarters, and to a new CR otherwise.20 19 Further 20 The
details on this dataset are provided in the Data Appendix. government also created new communes, which are technically non-rural. We code these commu-
16
We use New-cr to indicate villages assigned to a new CR in 2008, and Rump-cr to denote post-split villages assigned to a new CR under the old CR headquarters in 2008. Villages that do not fall into either category were not subject to a split in 2008. Specifically, we estimate the following regression:
transf ersijt = α + β1 Splitjt ∗ Rumpijt + β2 Splitjt ∗ N ewijt + δt + εijt
4.2
(2)
Targeting: Different Strategies for Different Groups?
In our second analysis we test the study’s main targeting hypotheses: that incumbents are less likely to target groups with weak brokers (H1a ), and that the history of reciprocal exchange conditions the targeting policies—public goods or new administrative units—that incumbents adopt toward groups with strong brokers (H1b and H1c ). Unit of analysis Early pioneering work on the determinants of administrative unit splits (e.g., Green (2010)), erroneously used the unit that split as unit of analysis.21 Later work conducted its analysis at one level below the unit that splits (e.g., Grossman and Lewis (2014); Hassan (2016)). While an improvement, this approach also suffers from problems since the boundaries of the cluster of villages that form new units are potentially endogenous. By using a unit that is stable over time—villages—we are able to control for selection into splitting status, as well as differential trends between splitting and stable units. Dependent variables Our main dependent variables are (1) assignment to a new CR, and (2) change in local and national public goods. For the first dependent variable, our analysis is restricted to the nities as rump or new CRs depending on whether they include the old CR headquarters. This coding is consistent with the decentralization law of 2014, which eliminated the distinction between rural and urban communities. 21 See comprehensive critique in Grossman and Lewis (2014, 200).
17
creation of 74 new CRs (from a baseline of 314, a 24% increase) that took place in 2008. Turning to the second set of dependent variables, we use ∆ Local Goods to measure changes in the provision of five locally administered public goods (clean water, primary schools, primary health centers, rural roads, and local markets) between 2000 and 2009. While local public goods are more excludable—and thus more likely to be targeted to specific areas—we also create the variable ∆National Goods using three goods administered at the national level: telephone networks, electricity, and paved roads. Public goods data are derived from village surveys conducted by the Senegalese National Statistics Agency in 2000 and 2009 and contain information about whether each type of public good is provided in each village. Using these data summarized in Appendix Table A.7, we create a local public goods index and a national public goods index for each year, which sum the binary access indicators for each set of public goods.
Independent Variables To test our differential targeting hypotheses, we group villages into three categories: Mouride, Toucouleur, and Other.22 We classify a village as Toucouleur or Mouride if over 50 percent of a village’s population share is reported to belong to that group according to the 2002 Senegalese census.23 In the one percent of villages in which a majority of the village is both Toucouleur and Mouride, we assign it “Other” status as our theoretical predictions are no longer clear. 22 “Other,”
the catch-all social grouping, serves as the omitted category in all regression analysis. Table A.1 in the Appendix introduces the population shares of the main ethnic and religious groups, as well as the corresponding shares for the grouping we use for this analysis. 23 We combine the Toucouleur with the Peul and Pulaar groups, as these self-reported census categories commonly overlap. Table A.15 in the Appendix shows robustness to alternatively considering only Toucouleur and Peul or only Toucouleur and Pulaar groups. Table A.16 in the Appendix indicate that our results are also robust to using different population share cutoff points (e.g., 60, 70 or 80 percent, rather than 50.)
18
Estimation We estimate how an incumbent’s distributive policy choices—the creation of a new CR, and the provision of national and local public goods—correlate with the social composition of villages. Formally, we fit the following model specification:
policyijt = α + β1 M ourideijt + β2 T oucouleurijt + εijt
(3)
where policyit is a variable indicating whether a specific policy was implemented in village i located in CR j (a dummy variable for CR creation and an integer for change in access to public goods), M ourideit indicates whether more than half of the village’s population self-identified as Mouride, T oucouleurit indicates whether more than half of the village’s population self-identified as Toucouleur.24 β1 and β2 are the two coefficients of interest; we cluster standard errors at the CR level. Building on the theoretical framework presented in Section 2, β1 is expected to be positive when examining change in public goods provision, since the Mouride have both a high coordinating capacity and a long history of reciprocal exchange with the incumbent. β2 is expected to be positive when examining new CR creation, since the Toucouleur have a high coordinating capacity, but no history of reciprocal exchange with the incumbent.
4.3
Electoral returns of the creation of new CRs
Finally, we estimate the electoral consequences of the creation of new CRs. Specifically, we test our hypothesis that areas that receive a new CR should exhibit a larger electoral support for the incumbent in the next election (H2 ), relative to all other areas. We also test that, within rump areas, those residing further from the CR headquarters (H3 ) should show a relatively larger support for the incumbent after being carved out. 24 We
do not include any controls since M ourideit and T oucouleurit are predetermined variables.
19
Dependent variables The key dependent variable in this final analysis, ∆Incumbent, measures the change in the vote share of Wade’s PDS, from the pre- to the post-split period. Electoral outcomes are measured at the polling station level and computed using data from Senegal’s Independent National Electoral Commission. Incumbent is defined as PDS vote share divided by the total number of valid votes. Past studies have argued that splits are designed to increase government presence at the grassroots level, and thereby increase the surveillance required to mobilize the electorate. Thus, we also construct a measure, Turnout, defined as the number of valid votes divided by the total number of registered voters. We use the change in this outcome variable to adjudicate between our hypotheses and this alternative explanation. Summary of these election outcomes can be found in the Appendix, Table A.6.
Independent variables Our key independent variables—New-cr and Rump-cr, defined above—capture changes in a village’s CR status over time. We focus on CR splits that took place in 2008, which represent the starkest and (empirically cleanest) episode of administrative-unit creation in Senegal’s recent history, directly affecting 1,656 of a total of 10,873 villages (Appendix, Table A.5).25 . Given that we are interested in the change in Wade’s PDS vote before and after 2008, we define a period indicator Post, which captures the second of the two major elections that took place in a relatively short timespan (2007 and 2009) around the 2008 splits. This strategy reduces the possibility that other confounding policies— for example, major public goods—took place at the village level between the creation of new CRs and the subsequent election. We also test robustness to comparing pre- and post-split elections for CR councilors (∆ from 2002 to 2009). 25 While
a few CRs (16) were also created between 2010 and 2011, these affected a relatively small number (318) of villages and they might have also led to increased public goods provision before the 2012 election
20
In order to test H3 —that the effects of splits on electoral outcomes depend on proximity to local government services—we further create a continuous measure of Distance to CR headquarters prior to 2008. To determine this distance, we use village geographical coordinates published by the Senegal National Statistics Agency, and fill in missing coordinates using data from Senegal’s ecological monitoring center. Since rural communities are named after their headquarters’ village or town, identifying the CR headquarters—to calculate the distance between a village and its old CR headquarters—is straightforward. We do not include measures of distance from a new CR headquarters in our estimation, since they are post-treatment.
Control variables Control variables, the purpose of which are described below, are constructed using Senegal’s 2002 census data, and include village population over the age of 20 (i.e., voting age eligibility in the 2000 elections); the population share of each major ethnicity and religious group; and household assets. Controls are first log-transformed, and then entered as linear, quadratic, and cubic variables. These variables are summarized in the Appendix, Tables A.8 and A.9.
Estimation We are interested in the causal effect of administrative unit creation, which is an endogenous distributive policy. To test this main implication of our theory while controlling for selection into administrative splits, we estimate the following difference-in-differences model:
yv,c,t = α0 + α1 dv,c + α2 nv,c dv,c + α3 rv,c dv,c + α4 nv,c + ηc + δpost + +β1 dv,c δpost + β2 nv,c dv,c δpost + β3 rv,c dv,c δpost + β4 nv,c δpost + ηc · δpost + +γ0 Xv,c + γ1 Xv,c δpost + εv,c,t
(4)
21
where the dependent variable yv,c,t is a measure of the incumbent’s vote share in village v, old CR c, and year t; dv,c is the village’s distance to the old CR headquarters; ηc is an indicator for the CR that the village v belonged to prior to 2008; nv,c is an indicator for New-cr villages; rv,c is an indicator for Rump-cr villages; δpost is an indicator for the post-2008 period; ηc is a fixed effect for the old CR; and Xv,c is a flexible vector of controls, described above.26 Note that for computational efficiency we run, and present estimates from Equation (4) in first differences. As a robustness check we also estimate this equation without the controls γ1 Xv,c .27 Our main coefficients of interest, β2 , β3 and β4 . β2 and β3 , capture heterogeneity in the effect of splits for New-cr and Rump-cr villages, respectively. Specifically, β3 > 0 implies that the magnitude of the positive effect of splits for villages in rump CRs increases with distance from the old CR headquarters, and β4 , which captures the changes in outcomes in New-cr villages compared to Rump-cr villages, provides a direct test of Hypothesis H3 . Equation (4) controls for selection into CR splits and for differential trends between split and stable CRs through the ηc and ηc · δpost terms, respectively. The other main effects and controls capture unobservable differences associated with distance from the old CR headquarters (in levels and over time), and differences associated with being inside a rump or a new CR (in levels and over time). Our main identification assumption is that, conditional on these controls, there are no differential trends in electoral outcomes within villages in split units, or across different distances from the old CR headquarters. We test this no differential trends assumption using electoral data prior to 2008, as described in Section 5. 26 The
omitted category in this specification is villages in non-split CRs, located at zero distance from the CR headquarters, in the pre-split period. Equation (4) deliberately omits terms that are collinear with the other fixed effects: in particular, rv,c is not included in levels since it is jointly collinear with nv,c and ηc . Similarly, rv,c · δpost is jointly collinear with ηc · δpost and nv,c · δpost ; rv,c · ηc and nv,c · ηc are collinear with the main effects rv,c , nv,c and ηc ; nv,c · ηc · δpost and rv,c · ηc · δpost are collinear with nv,c · δpost and rv,c · δpost . 27 Removing γ X 1 v,c from the baseline specification effectively removes γ1 Xv,c δpost . Robustness to this alternative specification suggests that results are unlikely to be driven by differential trends across villages that vary in dv,c , which could be correlated with Xv,c .
22
4.4
Electoral returns of the creation of public-goods targeting
This paper does not seek to estimate the electoral return of public-goods targeting for two reasons. First and foremost, contrary to the case of new CRs, local public-goods transfers are often distributed to groups that are bound in a reciprocal relationship with a particular party. Thus, transfers beget votes and votes beget transfers. Not only it is difficult to parse out whether goods are being targeted as a reward for past votes or a motivation for future votes, but we should also expect that increases in public-goods transfers to reciprocal groups may result in no over-time increases in electoral support for the incumbent because those groups are already voting for the incumbent at high rates. Secondly, the strategy that we employ to identify the effects of CR creation on changes in vote share cannot be similarly used for local public goods provision because the unit at which the policy is implemented (village) is the same as the unit of analysis for which we have only two periods of data, and consequently, we are unable to account for unit-specific trends while still having within-unit variation in targeted policies.
5
Results
In this section, we provide information on the study’s key findings.
5.1
Government Transfers
Adding up all transfers received by rump CRs and new CRs to their old-CR level, we find that CRs that split experience a large increase in total and per capita financial transfers from the national government after 2008. Figure 2 shows that, while CRs that split in 2008 received slightly fewer per capita fiscal transfers in 2007-2008 than those that did not, they received a significantly larger amount in the post-split period. Figure A.1 in the online appendix shows a similar pattern using instead logs of per capita transfers. Focusing within the CRs that split in 2008, we find that while both rump CRs and new CRs experienced a 23
jump in transfers right after the split relative to no-split CRs, only new CRs witnessed their significant increase sustained over time. Figure 2: The Effect of CR Splits on Per-capita Transfers (levels), 2007-2014
These graphical patterns are corroborated by the regression analysis formalized in equations (1) and (2) and shown in Table A.10 in the Appendix. We find that splits are associated with higher per capita transfers from the central government to CRs on the order of about 2,000 FCFA or around 50 % of the baseline mean. The magnitude of this effect is significantly larger for new CRs relative to rump CRs. While these estimates cannot be interpreted as causal, they provide strong suggestive evidence that splits are associated with larger transfers per capita in the long run. This finding is consistent with our argument that administrative unit creation provides incumbents with a tool to signal post-election commitment to voters.
5.2
Targeting
We now turn to testing our argument that incumbents target different policies to groups differing across the following two dimensions: (i) the coordinating capacity of brokers, and (ii) the history of reciprocal exchange with the incumbent. In Sections 2 and 3, we argued 24
that the Mouride constitute a group with both a high coordinating capacity and a long history of reciprocal exchange with the incumbent, and are thus less likely to be targeted with administrative splits. On the other hand, the Toucouleur are an example of a group with high coordinating capacity but no history of reciprocal relationship with the incumbent. They should therefore be targeted with administrative splits. Consistent with Hypothesis H1b , Table 1 shows that public goods are more likely targeted to areas dominated by the Mouride and less likely targeted to areas dominated by the Toucouleur. More so, consistent with Hypothesis H1c , splits are more likely to occur in areas dominated by the Toucouleur (though results fall just below reported significance levels) and significantly less likely to occur in areas dominated by the Mouride. These findings are thus additionally supportive of Hypothesis H1a —that both groups (because of their high coordinating capacity) should be targeted by at least one kind of policy relative to the omitted group. Table 1: Electoral Targeting of Mouride and Toucouleur/Peul/Pulaar Groups
Mouride (non-Touc/Peul/Pulaar)
Touc/Peul/Pulaar (non-Mouride)
(1) New CR -0.126∗∗∗ (0.028)
(2) Local goods 0.227∗∗ (0.082)
(3) National goods 0.299∗∗∗ (0.066)
0.072 (0.037)
-0.610∗∗∗ (0.068)
-0.371∗∗∗ (0.056)
(4) (5) New CR Local goods -0.050∗ 0.200∗∗ (0.024) (0.066) 0.030 (0.034)
-0.254∗∗∗ (0.058)
-0.139∗∗ (0.042)
0.386∗∗∗ (0.018)
Local Goods (2000)
0.489∗∗∗ (0.019)
National Goods (2000)
Observations Adjusted R2 Controls
(6) National goods 0.162∗∗ (0.049)
10873 0.048 No
10873 0.065 No
10873 0.060 No
10873 0.127 Yes
10873 0.280 Yes
10873 0.331 Yes
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged assets (linear, quadratic, cubic), and public goods (2000). ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
25
5.3
Elections
We now move to report our estimation of the effect of CR splits on the change in Wade’s vote share between 2007 and 2009. The first three columns in Table 2 are our baseline specifications (where distance is measured first in logs and then in levels), which include controls and fixed effects as discussed above. We then show robustness to removing controls. Consistent with Hypothesis H2 , throughout all specifications, New-cr villages are more likely to increase their support to the incumbent. This finding is consistent with our argument developed in Section 2.2 that voters have a strong preference to be granted a new administrative unit. Dealing with the dual concerns that anticipation effects might bias our estimates and that our estimates might be driven by comparing changes in Wade’s vote share between a national (2007) and a local election (2009), Table A.11 in the Appendix indicates that the results are robust to examining changes between the 2002 and 2009 local elections. We move to report our testing of Hypothesis H3 , which states that remote villages in rump areas will be more likely to reward the incumbent as compared to nearby villages since they place a high premium on administrative attention. Though we find that Rumpcr villages located further from the old CR headquarters exhibit increased support for the incumbent this effect is relatively small and significant only in a model without controls and where distance is measured in levels (Table 2, column 6).
Testing identification assumptions In Table A.2 we test some of our identification assumptions of the difference-in-difference estimation (equation 4). Most importantly, out results suggest that pre-split trends in incumbent support (between 2000 and 2007) across places that will split and places that will not are not significantly different. More so, the findings in Table A.2 suggests that the granting of new CRs are not a reward for past vote, a result that would be at odd with our argument that administrative unit splits are designed to lure voters and brokers from groups with which the incumbent does not have a history of reciprocal exchange. 26
Table 2: Effect of CR Creation on Incumbent Vote Share (2007 to 2009) (1) ∆ Inc.
(2) ∆ Inc. (log dist.) 0.127 (0.074)
(3) ∆ Inc. (dist.) 0.123∗ (0.054)
(5) (6) ∆ Inc. ∆ Inc. (log dist.) (dist.) 0.096∗ 0.152 0.129∗ (0.048) (0.080) (0.057)
Distance
0.006 (0.006)
0.000 (0.001)
0.002 (0.007)
-0.001 (0.001)
New CR (by 2009)=1 × Distance
0.006 (0.021)
0.001 (0.001)
-0.001 (0.025)
0.001 (0.002)
0.004 (0.002) -0.678∗∗∗ (0.026)
0.028 (0.016)
0.005∗ (0.002)
-0.677∗∗∗ (0.026)
0.022 (0.015) -0.678∗∗∗ (0.026)
4136 0.601 Yes
4136 0.601 Yes
4136 0.602 Yes
4136 0.425 No
4136 0.426 No
New CR (by 2009)=1
0.104∗ (0.046)
Rump CR (by 2009)=1 × Distance Incumbent (2007)
Observations Adjusted R2 Controls
(4) ∆ Inc.
4136 0.425 No
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged ethnic and religious group size (linear, quadratic, cubic), incumbent vote share in 2007, and logged assets (linear, quadratic, cubic). Fixed effects are entered at the old CR level. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
We have argued that incumbents grant new administrative units to groups that require a rather strong signal of commitment since pre-election promises are not credible when the group does not have a history of reciprocal exchange. To further probe the assumption that underlies this argument, we take advantage of the fact that some new CRs were only granted in 2010-2011, following the 2009 election. We then refit the model formalized in equation (4) this time using the latter splits as the key independent variables. Since these splits did not occur yet, we expect the effect of future splits to be insignificant. This expectation is borne out in our data, as reported in the Appendix, Table A.3.
Alternative explanations We turn to eliminate alternative explanations. First, rather than signaling greater administrative attention, the increased presence of the state might allow the ruling party to increase voter mobilization relative to the (possibly already high) baseline mobilization by brokers. In other words, a greater capacity for electoral mobilization—rather than changes in citizen
27
preferences—could be causing the increase in incumbent vote share in 2009. We test this alternative indirectly by examining the effect of administrative splits on voter turnout. A null effect of CR splits on turnout would suggest that the mobilization channel is not a serious concern. As Table A.4 in the Appendix makes clear, we find no discernible effect of CR creation on turnout. Villages in newly created CRs do not exhibit higher turnout—the coefficient for this variable is a precisely estimated zero. The same result is found for the interaction of New-cr and Rump-cr villages with distance from the old CR headquarters. These results suggest that this channel is unlikely to be a key mechanism. A second concern is that CR splits improved the ability of brokers to monitor voters by creating more homogeneous voting blocks, along religious or ethnic dimensions. To address this concern, we re-run our baseline specification interacting New-cr and Old-cr status with the ethnic and religious distance between each village and the average of its old CR. Table A.13 in the Appendix shows that our results are unlikely to be explained by possible homogenization of the new CR boundaries. First, the average effect of being a New-cr is robust to the inclusion of main effects for ethnic and religious fractionalization. Second, the interactions of New-cr and Old-cr with fractionalization yield mostly insignificant coefficients, and the only two significant coefficients have the opposite sign one would expect if homogenization were the driver of the increase in incumbent vote shares in new CRs. This is consistent with the fact that, as Table A.12 indicates, CR splits did not create administrative units that were substantially more homogeneous. Third, the results in Appendix Table A.13 also indicate that greater homogeneity in policy preferences in split CRs is unlikely to explain our main findings. A final concern is that CR splits followed demands of voters in areas that suffered political, economic and symbolic marginalization, which potentially exhibited increasing support for the incumbent. To deal with this concern, we test whether the creation of new CRs is predicted by baseline levels of local and national public goods, the ethnic and religious
28
distance between each village and the average of its CR, and an asset index and population, as well as the interactions of all these variables and the distance from the old CR headquarters. The largely null findings in Table A.14 in the Appendix indicate that our results are unlikely to be accounted explained by any of the mentioned marginalization categories. Overall, the estimates are not consistent with the alternative explanations discussed herein.
6
Conclusion
This paper advances a novel explanation for the rapid increase in the number of administrative splits across the developing world. Our theoretical argument is rooted in the context of new democracies, as well as many authoritarian regimes, in which incumbents are not free from the need to deliver in order to win increasingly competitive elections. Such countries are characterized by weak information environments and generally non-programmatic parties. In these context, voters use heuristics (such as ascriptive characteristics and elite cues) to infer candidates’ congruence with their preference for targeted benefits. In response, incumbents are increasingly using distributive policies strategically to signal their commitment, but face a problem that some targeted benefits may not emit a strong enough signal to lure groups with whom the incumbent party does not have a history of reciprocal exchange. We argue that incumbents adopt a policy of administrative units splits to target such groups, since the granting of a new administrative unit entails a relatively stable flow of central government transfers. By contrast, incumbents target public goods to groups that have strong brokers and a shared history of reciprocal exchange. We test these arguments using the case of Senegal and find robust support for our theoretical predictions. Our theory of differential targeting is most applicable to other low-income electoral regimes where two scope conditions hold: 1) access to information is poor such that politicians must signal congruence with voters through targeted transfers, and 2) brokers are available to facilitate the coordination of groups of voters as well as reciprocal exchange
29
relationships among some groups and parties. Our argument about why incumbents even sometimes choose administrative unit splits in spite of the potential retribution from voters and leaders in rump areas – because there are differential preferences within rump areas themselves – should travel to any democracy in which administrative unit creation is strategically enacted for electoral gain. Our signaling theory of distributive policy choice contributes to past work in several important ways. First, while, past studies—for example, Grossman and Lewis (2014), Hassan (2016), and Pierskalla (2016a)—also assume that electoral considerations dominate the strategic use of administrative-unit splits, they do not embed incumbent’s strategy within a larger framework of distributive policy choice. Second, our theory of strategic choice does not presuppose that incumbents are necessarily reactive to grassroots mobilization. Grossman and Lewis (2014) argue that incumbents mainly respond to bottom-up pressure, and that the demand for splits is strongest in areas that suffer political, economic, and symbolic marginalization. Similarly, Pierskalla (2016a) argues that national governments respond to demand from areas with higher capacity for collective action. This sort of reactive strategy may be relevant for countries (such as Uganda and Indonesia) where splits must be voted on first by the local government, but not in other contexts (such as Senegal and Kenya) where incumbents have close to full control over administrative unit splits.28 Third, our theoretical argument is not inconsistent with those arguing that the creation of new administrative units allows incumbents to strengthen patronage networks and co-opt local elites (Green, 2010). Using administrative unit splits to target groups that do not have a history of reciprocal exchange with the incumbent’s party can certainly help cement new alliances between the national government and local elites and brokers (Kimura, 2012). Yet a narrow focus on patronage jobs not only overlooks the benefits for local citizens, 28 Importantly,
our argument that incumbents target groups with strong brokers with high coordination capacity, is nonetheless consistent with the finding that groups’ with higher collective action capacity are more likely to be targeted (Pierskalla, 2016a).
30
but also sidesteps the fact that there are more efficient ways to target groups (that do not entail bloating the bureaucracy). Furthermore, our argument regrading the importance of administrative attention helps explain why voters in rump areas are unlikely to punish the incumbent for; a point that past theories have had a hard time explaining. While we explicitly argue that incumbents are more likely to target groups that have strong brokers, understanding the conditions that support brokers’ ability to coordinate votes is beyond the scope of this paper, offering exciting avenues for future work. Similarly, we argue that reciprocal exchange between societal groups and political party depends, in part, on the economic independence of brokers from the state. Future work should further explore the factors that sustain groups’ partisan bias even when parties are non-programmatic and non-ideological. From a policy perspective, the study offers a cautionary tale of how increased political competition may lead incumbents to adopt policies that may carry shortterm electoral gains, but arguably at the expense of longer term development goals.
References Arriola, Leonardo R, “Capital and opposition in Africa: Coalition building in multiethnic societies,” World Politics, 2013, 65 (02), 233–272. Baldwin, Kate, “Why vote with the chief? Political connections and public goods provision in Zambia,” American Journal of Political Science, 2013, 57 (4), 794–809. Beck, Linda J., Brokering democracy in Africa: The rise of clientelist democracy in Senegal, Palgrave Macmillan, 2008. Boas, Taylor C and F Daniel Hidalgo, “Controlling the airwaves: Incumbency advantage and community radio in Brazil,” American Journal of Political Science, 2011, 55 (4), 869– 885.
31
Boone, Catherine, “Decentralization as Political Strategy in West Africa,” Comparative Political Studies, 2003, 36 (4), 355–380. , Political topographies of the African state: Territorial authority and institutional choice, Cambridge: Cambridge University Press, 2003. Carlson, Elizabeth, “Ethnic Voting and Accountability in Africa: A Choice Experiment in Uganda,” World Politics, 2015, 67 (2), 353–385. Chandra, Kanchan, Why ethnic parties succeed: Patronage and ethnic head counts in India, Cambridge University Press, 2004. Clinton, Joshua D and Jason A Grissom, “Public information, public learning and public opinion: democratic accountability in education policy,” Journal of Public Policy, 2015, 35 (03), 355–385. Collier, Paul and Pedro C Vicente, “Violence, bribery, and fraud: the political economy of elections in Sub-Saharan Africa,” Public Choice, 2012, 153 (1-2), 117–147. Conroy-Krutz, Jeffrey, “Information and Ethnic Politics in Africa,” British Journal of Political Science, 2013, 43 (2), 345–373. Dahou, Tarik and Vincent Foucher, “Senegal since 2000 : rebuilding hegemony in a global age,” in Mustapha R. and L. Whitfield, eds., Africa since democratisation, London: J. Currey, 2009. Englebert, Pierre, “Pre-Colonial Institutions, Post-Colonial States, and Economic Development in Tropical Africa,” Political Research Quarterly, 2000, 53 (1), 7–36. Francois, Patrick, Ilia Rainer, and Francesco Trebbi, “How is power shared in Africa?,” Econometrica, 2015, 83 (2), 465–503.
32
Fujiwara, Thomas and Leonard Wantchekon, “Can Informed Public Deliberation Overcome Clientelism? Experimental Evidence from Benin,” American Economic Journal: Applied Economics, 2013, 5 (4), 241–255. Gingerich, Daniel W and Luis Fernando Medina, “The endurance and eclipse of the controlled vote: a formal model of vote brokerage under the secret ballot,” Economics & Politics, 2013, 25 (3), 453–480. Gottlieb, Jessica, “Explaining Variation in Broker Strategies: A Lab-in-the-Field Experiment in Senegal,” Unpublished manuscript, 2016. and Horacio Larreguy, “An Informational Theory of Electoral Targeting in Clientelistic Contexts: Evidence from Senegal,” Unpublished manuscript, 2016. Green, Elliott, “Patronage, District Creation, and Reform in Uganda,” Studies in Comparative International Development, 2010, 45 (1), 83–103. Grossman, Guy, “Renewalist Christianity and the Political Saliency of LGBTs: Theory and Evidence from Sub-Saharan Africa,” The Journal of Politics, 2015, 77 (2), 337–351. and Janet I Lewis, “Administrative Unit Proliferation,” American Political Science Review, 2014, 108 (1), 196–217. , Jan H Pierskalla, and Emma Boswell Dean, “Government Fragmentation and Public Goods Provision,” Journal of Politics, n.d. Harding, Robin and David Stasavage, “What democracy does (and doesn’t do) for basic services: School fees, school inputs, and African elections,” The Journal of Politics, 2014, 76 (01), 229–245. Hassan, Mai, “A State of Change: District Creation in Kenya After the Beginning of Multi-Party Elections,” Political Research Quarterly, 2016.
33
and Ryan Sheely, “Executive-Legislative Relations, Party Defections, and Lower-Level Administrative Unit Proliferation: Evidence from Kenya,” Comparative Political Studies, 2017. Kelly, Catherine L., “Here’s everything you need to know about Senegal’s recent referendum,” The Washington Post, April 3 2016. Key, Valdimer Orlando, The Responsible Electorate, Belknap Press of Harvard University Press, 1966. Kimura, Ehito, Political Change and Territoriality in Indonesia. Provincial Proliferation, Vol. 46, Routledge, 2012. Kitschelt, Herbert and Steven I. Wilkinson, “Citizen-politician linkages: an introduction,” in Herbert Kitschelt and Steven I. Wilkinson, eds., Patrons, clients, and policies: Patterns of democratic accountability and political competition, Cambridge: Cambridge University Press, 2007, pp. 1–49. Koter, Dominika, “King Makers: Local Leaders and Ethnic Politics in Africa,” World Politics, 2013, 65 (2), 187–232. Lewis, Janet I, “When Decentralization Leads to Recentralization: Subnational State Transformation in Uganda,” Regional & Federal Studies, 2014, 24 (5), 571–588. Lieberman, Evan S, Race and regionalism in the politics of taxation in Brazil and South Africa, Cambridge University Press, 2003. Lupu, Noam, Party Brands in Crisis: Partisanship, Brand Dilution, and the Breakdown of Political Parties in Latin America, Cambridge University Press, 2016. Malesky, Edmund, “Gerrymandering–Vietnamese style: escaping the partial reform equilibrium in a nondemocratic regime,” The Journal of Politics, 2009, 71 (1), 132–159.
34
Moehler, Devra C and Naunihal Singh, “Whose news do you trust? Explaining trust in private versus public media in Africa,” Political Research Quarterly, 2009. O’Brien, Donald B. Cruise, Saints and politicians: Essays in the organisation of a Senegalese peasant society, Cambridge: Cambridge University Press, 1975. Pierskalla, Jan H, “Splitting the Difference? The Politics of District Creation in Indonesia,” Comparative Politics, 2016, 48 (2), 249–268. , “The Proliferation of Decentralized Governing Units,” in E. Wibbels and J. Rodden, eds., Governance and the Future of Development Aid, Washington DC: USAID, 2016. Resnick, Danielle, “Continuity and change in Senegalese party politics: Lessons from the 2012 elections,” African Affairs, 2013, 112 (449), 623–645. , “Strategies of Subversion in Vertically-Divided Contexts: Decentralisation and Urban Service Selivery in Senegal,” Development Policy Review, 2014, 32 (S1), S61–S80. Riedl, Rachel Beatty, Authoritarian Origins of Democratic Party Systems in Africa, Cambridge University Press, 2014. Rueda, Miguel R, “Small Aggregates, Big Manipulation: Vote Buying Enforcement and Collective Monitoring,” American Journal of Political Science, 2016. Stokes, Susan, “Perverse Accountability: A Formal Model of Machine Politics with Evidence from Argentina,” American Political Science Review, 2005, 99(3), 315–325. Weghorst, Keith R and Staffan I Lindberg, “What Drives the Swing Voter in Africa?,” American Journal of Political Science, 2013, 57 (3), 717–734. Weitz-Shapiro, Rebecca, Curbing Clientelism in Argentina: Politics, Poverty, and Social Policy, Cambridge University Press, 2014.
35
Data Appendix We obtained the transfers data from the Direction des Collectivités Locales (DCL), a department of the Ministry of the Interior in charge of dealing with local government affairs. CRs receive two forms of transfers from the center: the fonds de dotation de la décentralisation (FDD), which are transfers designed to cover current expenditure, and the fonds d’équipement des collectivités locales (FECL), which are transfers designed to fund longerterm investment projects. Our measure of transfers aggregates these two variables. We obtained this data for the years 2007-2008, 2009-2012 and 2014. For 2007 we only have FDD data, and for 2008 we only have FECL data. For the sake of comparability we pooled together these two types of transfers across 2007-2008 and consider this the pre-split level of transfers across all CRs. The data on access to local and national public goods provision are from a public infrastructure survey of all rural villages in Senegal that was conducted in 2000 and 2009 by Senegal’s National Agency for Statistics and Demography. Access to local public goods is defined as follows: a village is coded as having access to water if there is a clean water source within 1km of the village; schools are accessible if they are located within 3km of the village; and health, unpaved roads, and markets are accessible if they are located within 5km of the village. Access to the three national goods are defined as existing within 5km of a village. Coding for water, school, and health are based on responses to several categories: water is coded as “1" if the village has access to drinking fountains, boreholes, or a well. School is coded based on whether the village has access to either primary schools or adult literacy centers. Finally, access to health is defined based on access to maternity wards, health posts, or clinics. Polling station-level election outcomes are from Senegal’s independent electoral commission (CENA). We use outcomes from the 2002, 2007 and 2009 elections. Election returns in 2007 capture the first and only round of votes for president, and are ideal because they are less subject to idiosyncrasies that arise from inter-party coalitions forming after the results 36
of the first round are announced. Election returns in 2002 and 2009 correspond to local elections. We focus on the vote share for PDS in 2002 and 2007, but in 2009 outcomes are defined based on the vote share of Sopi, Wade’s electoral bloc. Our census data comes from the 2002 Senegalese census —the RGPH (Recensement Général de la Population et de l’Habitat) 3—that was effectively conducted between 2000 and 2002. The rich census provides population statistics, household assets, and ethnic and religious affiliations, which comprise the key covariates for whose differential trends we control in our baseline specification. Specifically, we construct the following census controls: logged population (flexible), logged ethnic and religious group size (linear, quadratic, cubic), and logged assets (linear, quadratic, cubic).
37
Appendix with qualitative evidence on voter preferences for administrative splits Fieldwork we conducted in Senegal in Summer 2014 provides support in favor of our assumptions about voter preferences in Senegal. The example of the town of Sibassor illustrates the assumption of an increase in steady transfers: after a split had just been enacted, villagers “let out a sigh of relief” since the split was expected to yield “significant financial benefits” for the community (Sud Quotidien, December 9, 2010). Closely related, the preference of Senegalese citizens for a new administrative unit is related to increased control over resources. In Sibassor, residents of the carved-out unit expressed their desire for “more autonomy” in the management of local affairs and longed to “conduct political activities of their own” (Sud Quotidien, December 9, 2010). There is also ample evidence suggesting that CR splits are “sticky:” after Macky Sall won the 2012 elections, attempts by his government to undo several splits decided by former president Wade were met with widespread resistance. For example, in Bambilor, demonstrations were organized to oppose the plans of the new administration to undo a split enacted by Wade’s administration (Sud Quotidien, April 18, 2012). In support of the idea that the value of a new administrative unit increases the further one resides from the old local government’s headquarters, a Minister described administrative unit creation as “ the president’s answer to need to bring the administration closer to the administered” (Le Soleil, November 2010), while a major candidate for the 2012 presidential election “noted the anger of populations (. . . ) obliged to travel 100 kilometers to district headquarters” (Le Soleil, April 2010). As evidence of the potential costs of new administrative splits to citizens, in the town of Mbane, several local councilors went on a hunger strike to oppose an anticipated split, which they portrayed as “the unjustified and political erection of the village of [a] full Commune, without resources nor Hinterland, with the unstated goal of sidelining political adversaries in the management of the rural community” (Sud Quotidien, May 24, 2011). Finally, as 38
evidence that the balance of costs and benefits generally falls in favor of supporting splits, the local media regularly report on the satisfaction of local populations with recent splits (and their attribution of the split to the president’s actions). For example, following the split of Sangalkam, residents of the new CRs signed a joint declaration expressing their “unfailing support” and “engagement without reserve” for president Wade (Le Soleil, July 2011).
39
Figure and Table Appendix Figure A.1: Effect of Splits on Transfers (logs), 2007-2014
40
Table A.1: Share of Total Villages With Majority (1) All Villages
(2) New CR
(3) Old CR
Ethnicity Diola Manding Serer Toucouleur/Peul/Pulaar Wolof
0.034 0.037 0.105 0.364 0.377
0.034 0.070 0.019 0.547 0.191
0.022 0.051 0.108 0.452 0.250
Religion Khadrya Mouride Tidjane OtherM
0.112 0.291 0.505 0.020
0.120 0.071 0.678 0.039
0.092 0.190 0.619 0.032
0.278 0.353 0.338 10873
0.065 0.543 0.368 1656
0.176 0.439 0.331 1199
Panel A
Panel B Mouride (non-Toucouleur/Peul/Pulaar) Toucouleur/Peul/Pulaar (non-Mouride) Remaining categories Observations
Notes: Majority is defined as having greater than 50% of a village’s population share. Ethnicities and religions listed in this table include only groups that make up at least 2.5% of the total population in CRs that experienced redistricting in 2008.
41
Table A.2: Effect of Incumbent Vote Share (2000 to 2007) on Future CR Change (2) New CR (by 2009) 0.046 (0.040)
(3) New CR (by 2009) 0.069∗∗ (0.023)
Distance
0.061∗∗∗ (0.012)
∆ Incumbent × Distance
-0.017 (0.015) 3812 0.794 Yes
∆ Incumbent
Observations Adjusted R2 Controls
(1) New CR (by 2009) 0.005 (0.015)
3812 0.784 Yes
(4) New CR (by 2009) 0.015 (0.015)
(5) New CR (by 2009) 0.032 (0.040)
(6) New CR (by 2009) 0.072∗∗ (0.023)
0.007∗∗∗ (0.002)
0.056∗∗∗ (0.011)
0.007∗∗∗ (0.002)
-0.005∗∗∗ (0.002) 3812 0.797 Yes
-0.008 (0.016) 3812 0.777 No
-0.005∗∗ (0.002) 3812 0.781 No
3812 0.765 No
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged ethnic and religious group size (linear, quadratic, cubic), and logged assets (linear, quadratic, cubic). Fixed effects are entered at the old CR level. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
42
Table A.3: Effect of Future CR Creation on Incumbent Vote Share (2007 to 2009) (1) ∆ Incumbent New CR (post 2009)=1
0.031 (0.034)
(2) ∆ Incumbent (log distance) 0.004 (0.083)
(3) (4) ∆ Incumbent ∆ Incumbent (distance) 0.013 0.033 (0.053) (0.036)
(5) (6) ∆ Incumbent ∆ Incumbent (log distance) (distance) 0.098 0.061 (0.110) (0.060)
Distance
0.015∗ (0.006)
0.001 (0.001)
0.010 (0.007)
0.000 (0.001)
New CR (post 2009)=1 × Distance
0.008 (0.039)
0.002 (0.005)
-0.033 (0.041)
-0.003 (0.005)
0.002 (0.004) -0.678∗∗∗ (0.026)
-0.004 (0.023)
-0.001 (0.003)
-0.676∗∗∗ (0.026)
-0.000 (0.029) -0.677∗∗∗ (0.026)
4136 0.597 Yes
4136 0.598 Yes
4136 0.597 Yes
4136 0.422 No
4136 0.421 No
Rump CR (post 2009)=1 × Distance Incumbent (2007)
43
Observations Adjusted R2 Controls Notes: Same as in Table A.2.
4136 0.421 No
Table A.4: Effect of CR Creation on Turnout (2007 to 2009) (1) ∆ Turnout
(2) ∆ Turnout (log distance) -0.005 (0.044)
(3) ∆ Turnout (distance) -0.018 (0.022)
Distance
-0.001 (0.004)
-0.001 (0.000)
0.011∗∗ (0.004)
0.001 (0.001)
New CR (by 2009)=1 × Distance
-0.001 (0.016)
0.000 (0.001)
-0.003 (0.015)
-0.001 (0.001)
Rump CR (by 2009)=1 × Distance
0.000 (0.008)
-0.001 (0.001)
-0.006 (0.009)
-0.001 (0.001)
-0.617∗∗∗ (0.041) 4249 0.447 Yes
-0.620∗∗∗ (0.041) 4249 0.448 Yes
4249 0.312 No
4249 0.311 No
New CR (by 2009)=1
-0.010 (0.015)
-0.617∗∗∗ (0.041) 4249 0.448 Yes
Turnout (2007) Observations Adjusted R2 Controls
(4) ∆ Turnout 0.005 (0.018)
(5) (6) ∆ Turnout ∆ Turnout (log distance) (distance) -0.006 0.002 (0.044) (0.024)
4249 0.310 No
Notes: Same as in Table A.2.
Table A.5: Villages and CRs Affected by Split Year
Villages
2008 2010 2011 Total
1,656 144 174 1,974
Village Share 83.89% 7.29% 8.81% 100%
CRs 74 6 10 90
CR Population Share 82.22% 268,587 0.67% 19,992 0.11% 30,685 100% 319,264
44
Population Share 84.13% 6.26% 9.61% 100%
Table A.6: Election Summary Statistics count New CR (by 2009) 4635 Old CR (by 2009) 4635 New CR (post 2009) 4635 Old CR (post 2009) 4635 Distance 4635 ∆ Incumbent (2000-2009) 3631 ∆ Incumbent (2000-2007) 3812 ∆ Incumbent (2007-2009) 4136 ∆ Turnout (2000-2009) 3704 ∆ Turnout (2000-2007) 3834 Incumbent (2000) 3883 Incumbent (2007) 4383 Turnout (2000) 3889
mean sd 0.174 0.379 0.124 0.329 0.029 0.168 0.035 0.183 10.310 10.213 0.034 0.330 0.079 0.296 -0.044 0.269 -0.046 0.178 0.102 0.148 0.483 0.216 0.565 0.215 0.609 0.137
min max 0 1 0 1 0 1 0 1 0 111 -0.879 0.976 -0.880 0.941 -0.935 0.946 -0.931 0.738 -0.721 0.788 0 1 0.011 1 0 0.987
Table A.7: Public Goods Summary Statistics New CR (by 2009) Rump CR (by 2009) New CR (post 2009) Rump CR (post 2009) Distance ∆ Water ∆ School ∆ Health ∆ Road ∆ Paved Road ∆ Phone ∆ Electric ∆ Market ∆ Local Goods ∆ National Goods Local Goods (2000) Access to health (2000) Access to school (2000) Access to water (2000) Acess to local road (2000) Acess to market (2000) National Goods (2000) Paved Road (2000) Electric (2000) Phone (2000)
count mean sd min max 10873 0.152 0.359 0 1 10873 0.110 0.313 0 1 10873 0.029 0.169 0 1 10873 0.035 0.184 0 1 10873 10.010 9.830 0 111 10873 0.229 0.559 -1 1 10873 0.158 0.507 -1 1 10873 0.141 0.511 -1 1 10873 0.164 0.551 -1 1 10873 0.071 0.414 -1 1 10873 0.186 0.536 -1 1 10873 0.074 0.475 -1 1 10873 0.044 0.439 -1 1 10873 0.737 1.414 -5 5 10873 0.332 1.025 -3 3 10873 2.106 1.431 0 5 10873 0.365 0.481 0 1 10873 0.688 0.463 0 1 10873 0.530 0.499 0 1 10873 0.310 0.462 0 1 10873 0.213 0.410 0 1 10873 0.768 1.060 0 3 10873 0.286 0.452 0 1 10873 0.169 0.375 0 1 10873 0.313 0.464 0 1
45
Table A.8: Ethnicity Summary Statistics Population Share Badiaran Share Bainouk Share Balante Share Bambara Share Bassari Share Bedick Share Coniagui Share Creole Share Diakhank Share Dialonke Share Diola Share Foreigner Share Fula Share Laobe Share Lebou Share Malinke Share Mancagne Share Manding Share Manjag Share Maure Share OtherE Share Peul Share Pulaar Share Sarakole Share Serer Share Soce Share Soninke Share Soussou Share Tandanke Share Toucoule Share Wolof Share Catholic Share Khadrya Share Layenne Share Mouride Share OtherC Share OtherM Share OtherR Share Protestant Share Tidjane
count mean sd 10873 195.110 601.029 10873 0.001 0.018 10873 0.001 0.023 10873 0.009 0.073 10873 0.010 0.062 10873 0.002 0.044 10873 0.000 0.018 10873 0.001 0.011 10873 0.000 0.001 10873 0.003 0.043 10873 0.002 0.038 10873 0.033 0.167 10873 0.002 0.017 10873 0.000 0.003 10873 0.002 0.021 10873 0.000 0.013 10873 0.002 0.035 10873 0.002 0.037 10873 0.042 0.165 10873 0.007 0.058 10873 0.009 0.069 10873 0.000 0.010 10873 0.307 0.406 10873 0.060 0.203 10873 0.004 0.043 10873 0.112 0.274 10873 0.002 0.025 10873 0.005 0.053 10873 0.000 0.002 10873 0.000 0.005 10873 0.017 0.093 10873 0.364 0.430 10873 0.025 0.113 10873 0.131 0.261 10873 0.002 0.026 10873 0.299 0.382 10873 0.002 0.026 10873 0.031 0.122 10873 0.006 0.052 10873 0.000 0.008 10873 0.505 0.405
min max 1 37138 0 1.000 0 0.876 0 1.000 0 1.000 0 1.000 0 1.000 0 0.643 0 0.088 0 0.982 0 1.000 0 1.000 0 0.750 0 0.237 0 1.000 0 0.788 0 1.000 0 1.000 0 1.000 0 1.000 0 1.000 0 0.768 0 1.000 0 1.000 0 1.000 0 1.000 0 0.909 0 1.000 0 0.150 0 0.362 0 1.000 0 1.000 0 1.000 0 1.000 0 1.000 0 1.000 0 0.951 0 1.000 0 1.000 0 0.377 0 1.000
Population Share Badiaran Share Bainouk Share Balante Share Bambara Share Bassari Share Bedick Share Coniagui Share Creole Share Diakhank Share Dialonke Share Diola Share Foreigner Share Fula Share Laobe Share Lebou Share Malinke Share Mancagne Share Manding Share Manjag Share Maure Share OtherE Share Pulaar Share Sarakole Share Serer Share Soce Share Soninke Share Soussou Share Tandanke Share Wolof Share Catholic Share Khadrya Share Layenne Share Mouride Share OtherC Share OtherM Share OtherR Share Protestant Share Tidjane
46
count 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635 4635
mean sd 309.138 393.738 0.001 0.014 0.002 0.030 0.008 0.065 0.009 0.056 0.002 0.034 0.001 0.022 0.001 0.010 0 0.001 0.006 0.056 0.003 0.043 0.044 0.189 0.002 0.015 0 0.004 0.002 0.013 0.001 0.013 0.003 0.044 0.001 0.027 0.063 0.202 0.005 0.043 0.008 0.061 0.001 0.015 0.068 0.213 0.005 0.046 0.153 0.318 0.002 0.026 0.009 0.076 0 0.002 0 0.007 0.342 0.418 0.029 0.111 0.137 0.265 0.002 0.022 0.277 0.358 0.002 0.025 0.035 0.127 0.007 0.054 0.001 0.012 0.510 0.392
min max 2 10805 0 0.777 0 1 0 1 0 1 0 0.990 0 1 0 0.384 0 0.088 0 0.950 0 0.995 0 1 0 0.473 0 0.237 0 0.383 0 0.788 0 1 0 1 0 1 0 0.983 0 1 0 0.768 0 1 0 1 0 1 0 0.909 0 1 0 0.050 0 0.362 0 1 0 1 0 1 0 1 0 1 0 0.946 0 1 0 0.960 0 0.377 0 1
Table A.9: Assets Summary Statistics Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share
radio television video refrigerator telephone stove_cooking fireplace airconditioner sewingmachine car moped bicycle carriage pirogue hoe cart milkinganimals tractor truck mopedbike pirogue2 refrigerator2 sewingmachine2 musicequip chair fax photocopier computer mill camera building
count 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873 10873
mean 0.792 0.064 0.009 0.008 0.017 0.021 0.012 0.001 0.011 0.024 0.041 0.167 0.441 0.008 0.693 0.367 0.463 0.006 0.007 0.010 0.006 0.003 0.008 0.002 0.006 0.002 0.000 0.000 0.007 0.002 0.014
sd 0.200 0.117 0.042 0.039 0.064 0.085 0.074 0.012 0.038 0.075 0.088 0.284 0.296 0.045 0.326 0.301 0.359 0.040 0.031 0.043 0.042 0.019 0.034 0.022 0.039 0.016 0.009 0.009 0.043 0.019 0.083
min 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
max 1.000 1.000 0.842 1.000 0.830 1.000 1.000 0.830 1.000 1.000 1.000 1.000 1.000 0.861 1.000 1.000 1.000 1.000 1.000 1.000 0.871 0.830 1.000 1.000 1.000 0.830 0.830 0.830 1.000 0.830 1.000
Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share Share
47
radio television video refrigerator telephone cooking stove fireplace air conditioner sewing machine car moped bicycle carriage pirogue hoe cart milking animals tractor truck moped bike pirogue 2 refrigerator 2 sewing machine 2 music equipment chair fax photocopier computer mill camera building
count mean sd min max 4635 0.775 0.181 0 1 4635 0.076 0.114 0 1 4635 0.012 0.041 0 0.825 4635 0.013 0.043 0 0.825 4635 0.020 0.062 0 0.825 4635 0.023 0.079 0 1 4635 0.012 0.071 0 1 4635 0.001 0.016 0 0.825 4635 0.015 0.036 0 0.825 4635 0.027 0.068 0 1 4635 0.045 0.080 0 0.867 4635 0.167 0.270 0 1 4635 0.415 0.274 0 1 4635 0.011 0.050 0 0.779 4635 0.678 0.311 0 1 4635 0.345 0.276 0 1 4635 0.440 0.339 0 1 4635 0.006 0.034 0 0.866 4635 0.008 0.032 0 1 4635 0.011 0.040 0 0.825 4635 0.009 0.055 0 0.868 4635 0.005 0.023 0 0.825 4635 0.011 0.036 0 1 4635 0.003 0.028 0 1 4635 0.007 0.035 0 0.825 4635 0.002 0.019 0 0.825 4635 0 0.014 0 0.825 4635 0 0.014 0 0.825 4635 0.006 0.033 0 1 4635 0.002 0.019 0 0.825 4635 0.018 0.093 0 1
Table A.10: Effect of CR Split on Per Capita Transfers to CRs Level CR split
(1) 2029.437∗∗∗ [529.699]
CR Split*Rump CR Split*New Mean Observations Adjusted R2
4318.053 2040 0.09
Logarithmic (2)
799.712 [569.064] 3564.163∗∗∗ [783.038] 4318.053 2040 0.12
(3) 0.349∗∗∗ [0.078]
8.011 2040 0.20
(4)
0.173∗∗ [0.088] 0.615∗∗∗ [0.096] 8.011 2040 0.24
Notes: Total per capita transfers’ denominator is in FCFA and the denominator is the 2002 population. Robust standard errors in parentheses, clustered at the old CR level. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
48
Table A.11: Effect of CR Creation on Incumbent Vote Share (2002 to 2009) (1) ∆ Incumbent New CR (by 2009)=1
0.096∗ (0.041)
(2) ∆ Incumbent (log distance) 0.095 (0.073)
(3) (4) ∆ Incumbent ∆ Incumbent (distance) 0.107∗ 0.093 (0.049) (0.048)
(5) (6) ∆ Incumbent ∆ Incumbent (log distance) (distance) 0.104 0.106 (0.090) (0.058)
Distance
0.007 (0.007)
0.000 (0.001)
0.001 (0.009)
-0.001 (0.001)
New CR (by 2009)=1 × Distance
0.011 (0.021)
0.001 (0.001)
0.006 (0.028)
0.002 (0.002)
Rump CR (by 2009)=1 × Distance
0.018 (0.015)
0.004 (0.002)
0.014 (0.019)
0.004 (0.003)
-0.868∗∗∗ (0.025) 3552 0.689 Yes
-0.867∗∗∗ (0.025) 3552 0.689 Yes
3552 0.411 No
3552 0.412 No
Incumbent (2002)
49
Observations Adjusted R2 Controls
-0.866∗∗∗ (0.025) 3552 0.688 Yes
3552 0.411 No
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged ethnic and religious group size (linear, quadratic, cubic), incumbent vote share in 2007, and logged assets (linear, quadratic, cubic). Fixed effects are entered at the old CR level. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A.12: Effect of CR Creation on Incumbent Vote Share (Religion and Ethnic Distance)
New CR (by 2009)=1
(1) (2) (3) (4) (5) (6) ∆ Incumbent ∆ Incumbent ∆ Incumbent ∆ Incumbent ∆ Incumbent ∆ Incumbent 0.104∗ 0.160∗∗ 0.138∗ 0.096∗ 0.160∗ 0.157∗ (0.046) (0.061) (0.067) (0.048) (0.067) (0.071)
Religious distance
0.023 (0.013)
0.021 (0.014)
New CR (by 2009)=1 × Religious distance
-0.070∗ (0.034)
-0.101∗ (0.041)
Old CR (by 2009)=1 × Religious distance
0.022 (0.035)
0.001 (0.039) -0.018 (0.014)
-0.032∗ (0.015)
New CR (by 2009)=1 × Ethnic distance
-0.036 (0.030)
-0.062 (0.041)
Old CR (by 2009)=1 × Ethnic distance
0.014 (0.028)
0.026 (0.037)
50
Ethnic distance
Incumbent (2007)
-0.677∗∗∗ (0.026)
-0.676∗∗∗ (0.026)
-0.675∗∗∗ (0.026)
Constant
0.373∗∗∗ (0.071) 4136 0.601 Yes
0.353∗∗∗ (0.071) 4136 0.602 Yes
0.385∗∗∗ (0.072) 4136 0.601 Yes
Observations Adjusted R2 Controls
-0.058∗∗∗ (0.007) 4136 0.425 No
-0.070∗∗∗ (0.011) 4136 0.426 No
-0.045∗∗∗ (0.012) 4136 0.428 No
Notes: Ethnic and religious distance measures are the absolute value of the difference between 1) each ethnic/religious group’s population share within the village; and 2) the weighted mean population share of that ethnic/religious group in the CR. Standard errors and controls are the same as in Table A.2. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A.13: Average Ethnic and Religious Herfindahl Ethnic HHI
Religious HHI
0.550 0.590 0.562
0.592 0.614 0.593
0.000 0.021
0.000 0.009
CRs before split New CRs after split Rump CRs after split T-test (p-values) New CR 6= before split Rump CR 6= before split
Notes: Ethnicity and religion categories used are same as those in Table 1.
51
Table A.14: Effect of Non-Political Factors on Future CR Change (1) New CR (by 2009)
(2) New CR (by 2009) (log distance) 0.085∗∗∗ (0.025)
(3) New CR (by 2009) (distance) 0.005∗∗ (0.002)
-0.003 (0.002)
0.029∗ (0.012)
0.003 (0.004)
-0.010 (0.006)
0.000 (0.000)
Distance
Local Goods (2000) Local Goods (2000) × Distance
National Goods (2000)
(4) New CR (by 2009)
(5) New CR (by 2009) (log distance) 0.077∗∗∗ (0.019)
(6) New CR (by 2009) (distance) 0.005∗∗ (0.002)
-0.007∗∗ (0.002)
0.046∗∗ (0.016)
0.003 (0.006)
-0.020∗∗ (0.007) 10873 0.793 No
0.000 (0.001) 10873 0.793 No
National Goods (2000) × Distance Observations Adjusted R2 Controls
10873 0.782 No
10873 0.792 No
10873 0.793 No
10873 0.782 No
(1) New CR (by 2009)
(2) New CR (by 2009) (log distance) 0.045∗∗ (0.014)
(3) New CR (by 2009) (distance) 0.004∗ (0.002)
(4) New CR (by 2009)
(5) New CR (by 2009) (log distance) 0.043∗∗∗ (0.013)
(6) New CR (by 2009) (distance) 0.004∗ (0.002)
-0.004 (0.007)
-0.048 (0.040)
-0.029 (0.018)
0.016 (0.020)
0.002 (0.002) -0.007 (0.006)
-0.054 (0.035)
-0.035∗ (0.017)
0.019 (0.017)
0.002 (0.002)
Distance
Religious distance Religious distance × Distance
Ethnic distance Ethnic distance × Distance
Observations Adjusted R2 Controls
10873 0.782 No
10873 0.791 No
10873 0.794 No
10873 0.782 No
10873 0.791 No
10873 0.794 No
(1) New CR (by 2009)
(2) New CR (by 2009) (log distance) 0.055∗∗∗ (0.012)
(3) New CR (by 2009) (distance) 0.005∗∗∗ (0.002)
(4) New CR (by 2009)
(5) New CR (by 2009) (log distance) 0.083∗∗ (0.026)
(6) New CR (by 2009) (distance) 0.003 (0.002)
0.000 (0.002)
0.012 (0.007)
0.003 (0.003)
-0.003 (0.005)
0.001 (0.001) -0.000 (0.002)
0.017∗ (0.008)
-0.001 (0.004)
-0.006 (0.004)
0.000 (0.000)
10873 0.791 No
10873 0.793 No
Distance
Wealth Wealth × Distance
Population (log) Population (log) × Distance
Observations Adjusted R2 Controls
10873 0.782 No
10873 0.791 No
10873 0.793 No
10873 0.782 No
Notes: Ethnic and religious distance measures are the absolute value of the difference between 1) each ethnic/religious group’s population share within the village; and 2) the weighted mean population share of that ethnic/religious group in the CR. Robust standard errors in parentheses, clustered at the old CR level. Fixed effects are entered at the old CR level. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
52
Table A.15: Effect of Mouride and Toucouleur Groups (Alternative Definitions)
Mouride (non-Touc/Peul)
Touc/Peul (non-Mouride)
(1) (2) New CR (by 2009) Local goods (2009) ∗∗∗ -0.141 0.290∗∗∗ (0.027) (0.081) 0.051 (0.038)
(3) National goods (2009) 0.334∗∗∗ (0.065)
(4) New CR (by 2009) -0.065∗∗ (0.023)
(5) Local goods (2009) 0.246∗∗∗ (0.065)
(6) National goods (2009) 0.184∗∗∗ (0.049)
-0.356∗∗∗ (0.054)
-0.004 (0.032)
-0.167∗∗ (0.058)
-0.117∗∗ (0.042)
-0.553∗∗∗ (0.063)
0.388∗∗∗ (0.019)
Local Goods (2000)
0.490∗∗∗ (0.019)
National Goods (2000)
Constant Observations Adjusted R2 Controls
Mouride (non-Touc/Pulaar)
Touc/Pulaar (non-Mouride)
0.177∗∗∗ (0.028) 10873 0.044 No
2.928∗∗∗ (0.047) 10873 0.058 No
(1) (2) New CR (by 2009) Local goods (2009) -0.153∗∗∗ 0.501∗∗∗ (0.027) (0.082) 0.091 (0.065)
1.113∗∗∗ (0.044) 10873 0.058 No
0.213∗∗∗ (0.063) 10873 0.126 Yes
1.229∗∗∗ (0.195) 10873 0.277 Yes
0.517∗∗ (0.156) 10873 0.330 Yes
(3) National goods (2009) 0.476∗∗∗ (0.066)
(4) New CR (by 2009) -0.050∗∗ (0.019)
(5) Local goods (2009) 0.281∗∗∗ (0.064)
(6) National goods (2009) 0.216∗∗∗ (0.048)
-0.069 (0.071)
0.121∗ (0.061)
-0.277∗∗∗ (0.078)
-0.066 (0.048)
-0.226 (0.121)
0.392∗∗∗ (0.018)
Local Goods (2000)
0.493∗∗∗ (0.019)
National Goods (2000)
Constant Observations Adjusted R2 Controls
0.191∗∗∗ (0.028) 10873 0.044 No
2.714∗∗∗ (0.046) 10873 0.031 No
0.967∗∗∗ (0.042) 10873 0.040 No
0.210∗∗ (0.066) 10873 0.131 Yes
1.157∗∗∗ (0.194) 10873 0.277 Yes
0.470∗∗ (0.158) 10873 0.329 Yes
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged assets (linear, quadratic, cubic), and public goods (2000). Panel 1 defines Toucouleur as Toucouleur or Peul, and panel 2 defines Toucouleur as Toucouleur or Pulaar. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
53
Table A.16: Electoral Targeting of Mouride and Toucouleur/Peul/Pulaar Groups (Alternative Cutoffs) Cutoff = 60% Mouride (non-Touc/Peul/Pulaar)
Touc/Peul/Pulaar (non-Mouride)
(1) (2) New CR (by 2009) Local goods (2009) -0.123∗∗∗ 0.229∗∗ (0.026) (0.083) 0.078∗ (0.037)
-0.630∗∗∗ (0.069)
(3) National goods (2009) 0.317∗∗∗ (0.065)
(4) New CR (by 2009) -0.050∗ (0.022)
-0.380∗∗∗ (0.057)
0.030 (0.033)
(5) (6) Local goods (2009) National goods (2009) 0.206∗∗ 0.178∗∗∗ (0.066) (0.048) -0.268∗∗∗ (0.058) 0.385∗∗∗ (0.018)
Local Goods (2000)
0.489∗∗∗ (0.019)
National Goods (2000)
Constant Observations Adjusted R2 Controls
Cutoff = 70% Mouride (non-Touc/Peul/Pulaar)
Touc/Peul/Pulaar (non-Mouride)
0.157∗∗∗ (0.027) 10873 0.045 No
2.999∗∗∗ (0.048) 10873 0.065 No
(1) (2) New CR (by 2009) Local goods (2009) -0.117∗∗∗ 0.201∗ (0.024) (0.085) 0.088∗ (0.036)
-0.680∗∗∗ (0.069)
1.149∗∗∗ (0.048) 10873 0.060 No
0.197∗∗ (0.064) 10873 0.126 Yes
(3) National goods (2009) 0.297∗∗∗ (0.065)
(4) New CR (by 2009) -0.045∗ (0.020)
-0.423∗∗∗ (0.056)
0.035 (0.033)
1.267∗∗∗ (0.192) 10873 0.281 Yes
(5) (6) Local goods (2009) National goods (2009) 0.189∗∗ 0.171∗∗∗ (0.068) (0.047) -0.311∗∗∗ (0.058)
Cutoff = 80% Mouride (non-Touc/Peul/Pulaar)
Touc/Peul/Pulaar (non-Mouride)
0.149∗∗∗ (0.026) 10873 0.042 No
3.016∗∗∗ (0.046) 10873 0.067 No
(1) (2) New CR (by 2009) Local goods (2009) ∗∗∗ -0.116 0.190∗ (0.023) (0.091) 0.096∗∗ (0.036)
-0.691∗∗∗ (0.070)
1.170∗∗∗ (0.047) 10873 0.059 No
0.193∗∗ (0.064) 10873 0.126 Yes
(3) National goods (2009) 0.312∗∗∗ (0.067)
(4) New CR (by 2009) -0.046∗ (0.019)
-0.436∗∗∗ (0.055)
0.037 (0.032)
1.292∗∗∗ (0.192) 10873 0.282 Yes
(5) (6) Local goods (2009) National goods (2009) ∗ 0.186 0.186∗∗∗ (0.074) (0.049) -0.319∗∗∗ (0.056)
-0.164∗∗∗ (0.041)
0.489∗∗∗ (0.019)
National Goods (2000)
Observations Adjusted R2 Controls
0.538∗∗∗ (0.156) 10873 0.331 Yes
0.386∗∗∗ (0.018)
Local Goods (2000)
Constant
-0.162∗∗∗ (0.043)
0.488∗∗∗ (0.019)
National Goods (2000)
Observations Adjusted R2 Controls
0.525∗∗∗ (0.156) 10873 0.331 Yes
0.384∗∗∗ (0.018)
Local Goods (2000)
Constant
-0.140∗∗ (0.042)
0.145∗∗∗ (0.024) 10873 0.040 No
3.015∗∗∗ (0.045) 10873 0.064 No
1.173∗∗∗ (0.046) 10873 0.058 No
0.190∗∗ (0.064) 10873 0.126 Yes
1.307∗∗∗ (0.192) 10873 0.281 Yes
0.542∗∗∗ (0.157) 10873 0.331 Yes
Notes: Robust standard errors in parentheses, clustered at the old CR level. Included controls are logged population (flexible), logged assets (linear, quadratic, cubic), and public goods (2000). Panel 1 defines the cutoff at the plurality of population share, while panels 2, 3, and 4 define the cutoffs at 60%, 70%, and 80% of the population share, respectively. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
54
Table A.17: List of CR Splits Department (2002)
CR (2002)
Rump CR
New CR
Diourbel Diourbel Diourbel Fatick Fatick Fatick
Mbacke Mbacke Mbacke Fatick Foundiougne Foundiougne
Dendeye Gouygui Ndioumane Sadio Diakhao Diossong Djilor
Dendeye Gouygui (2008) Ndioumane (2008) Sadio (2006) Diakhao (c) (2011) Diossong (2010) Djilor (2008)
Fatick Fatick Fatick
Foundiougne Gossas Gossas
Darou Nahim (2008) Taiba Thiekene (2008) Khelcom (2006) Thiaré Ndialgui (2011) Diagane Barka, Niass‘ene (2010) Soum (c) (2008) Diagane Barka (2010) Mbam (2011) Karang Poste (c) (2008) Khelcom (2006) Nguelou, Khelcom-Birane (2008)
Fatick Fatick Fatick Fatick Kaolack
Gossas Gossas Gossas Gossas Kaffrine
Kaolack Kaolack Kaolack Kaolack Kaolack Kaolack
Kaffrine Kaffrine Kaffrine Kaffrine Kaffrine Kaffrine
Kaolack Kaolack Kaolack
Kaffrine Kaffrine Kaffrine
Kaolack Kaolack
Kaolack Nioro
Kaolack
Nioro
Kolda
Kolda
Kolda Kolda Kolda
Kolda Kolda Kolda
Kolda
Kolda
55
Region (2002)
Decree
Decret Decret Decret Decret Decret Decret Decret Decret Keur Samba Gueye Keur Samba Gueye (2008) Decret Colobane Colobane (2006) Decret Gagnick Gagnick (2008) Decret Decret Mbadakhoune Mbadakhoune (2008) Khelcom-Birane (2008) Decret Mbar Mbar (2008) Nguelou (2008) Decret Ndiago Ndiago (2008) Nguelou (2008) Decret Ourour Ourour (2008) Khelcom-Birane (2008) Decret Birkelane Birkelane (c) (2008) Touba Mbella, Keur Mboucki (2008) Decret Decret Diamal (2011) Decret Darou Miname Darou Miname (2006) Khelcom (2006) Decret Gniby Gniby (2006) Khelcom (2006) Decret Ida Mouride Ida Mouride (2008) Fass Thiekene (2008) Decret Mabo Mabo (2011) Ségré-Gatta, Mbeuleup (2011) Decret Maka Yop Maka Yop (2008) Missirah Wad‘ene (2008) Decret Maleme Hoddar Maleme Hoddar (c) (2008) Sagna (2008) Decret Decret Mboss Mboss (c) (2011) Dara Mboss, Panal Wolof (2011) Decret Ndioum Nguinthe Ndioum Nguinthe (2011) Ndiob‘ene Samba Lama (2011) Decret Nganda Nganda (c) (2008) Diamagadio (2008) Decret Decret Dya Dya (2010) Sibassor (c) (2010) Decret Keur Madiabel Keur Madiabel (c) (2008) Keur Madongo (2008) Decret Decret Paoskoto Paoskoto (2008) Keur Madiabel (c) (2008) Decret Dabaly, Darou Salam (2010) Decret Dabo Dabo (c) (2008) Dialambere (2008) Decret Decret Fafacourou Fafacourou (2008) Badion (2008) Decret Mampatim Mampatim (2008) Madina Cheriff (2008) Decret Medina Yoro Foulah Medina Yoro Foulah (c) (2008) Dinguiraye, Niaming (2008) Decret Decret Ndorna Ndorna (2008) Bignarabe, Bourouco, Koulinto (2008) Decret
no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no no
2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2006-391 du 27 avril 2006 2011-426 du 29 mars 2011 2010-1541 du 29 novembre 2010 2008-748 du 10 juillet 2008 2010-1541 du 29 novembre 2010 2011-428 du mars 2011 2008-748 du 10 juillet 2008 2006-391 du 27 avril 2006 2008-749 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2011-423 du 29 mars 2011 2006-391 du 27 avril 2006 2006-391 du 27 avril 2006 2008-749 du 10 juillet 2008 2011-423 du 29 mars 2011 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2011-431 du 29 mars 2011 2011-430 du 29 mars 2011 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2010-1543 du 29 novembre 2010 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2010-1542 du 29 novembre 2010 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-749 du 10 juillet 2008
56
Region (2002)
Department (2002)
CR (2002)
Rump CR
New CR
Kolda
Kolda
Pata
Pata (c) (2008)
Kérevane (2008)
Kolda
Kolda
Kolda Kolda Kolda Kolda
Kolda Kolda Sedhiou Sedhiou
Kolda
Sedhiou
Kolda Kolda
Sedhiou Sedhiou
Kolda
Sedhiou
Kolda Kolda
Sedhiou Sedhiou
Kolda
Sedhiou
Kolda
Sedhiou
Kolda
Velingara
Kolda Louga
Velingara Kebemer
Louga
Kebemer
Louga
Linguere
Louga Louga Matam Matam
Linguere Louga Kanel Kanel
Matam Saint Louis
Matam Dagana
Decree
Decret Decret Salikegne Salikegne (c) (2008) Guiro Yero Bocar (2008) Decret Decret Sare Bidji Sare Bidji (2008) Thietty (2008) Decret Tankanto Escale Tankanto Escale (2008) Sare Yoba Diega (c) (2008) Decret Bona Bona (2008) Diacounda (2008) Decret Bounkiling Bounkiling (c) (2008) Inor, Kandion Mangana Decret Madina Wandifa (c) (2008) Decret Djinany (2010) Decret Diannah Malari Diannah Malari (c) (2008) Diannah Ba, Sama Kanta Peulh (2008) Decret Decret Diaroume Diaroume (2008) Diambati, Faoune (2008) Decret Diattacounda Diattacounda (c) (2008) Simbandi Balante (2008) Decret Decret Diende Diende (2008) Koussy (2008) Decret Decret Djibanar Djibanar (2008) Kaour (2008) Decret Sakar Sakar (2008) Oudoucar (2008) Decret Decret Samine Escale Samine Escale (c) (2008) Mangouroungou Santo Decret Yarang Banlante (2008) Decret Tanaff Tanaff (c) (2008) Baghere, Diouboudou (2008) Decret Decret Decret Kounkane Kounkane (c) (2008) Diaobé (c), Kandiaye (2008) Decret Decret Paroumba Paroumba (2008) Pakour (2008) Decret Darou Marnane Darou Marnane (2008) Mbacke Kadior (2008) Decret Decret Gueoul Gueoul (c) (2008) Ngourane Ouolof (2008) Decret Decret Mbeuleukhe Mbeuleukhe (c) (2010) Yang Yang (2010), Yang Yang (2011) Decret Decret Sagatta Djolof Sagatta Djolof (2011) Affe Wolof (2011) Decret Ndiagne Ndiagne (c) (2011) Guet Ardo (2011) Decret Bokiladji Bokiladji (2008) Dembankane (c) (2008) Decret Sinthiou Bamanbe Sinthiou Bamanbe (c) (2008) Hamady Ounare (c), Ndendory (2008) Decret Decret Bokidiawe Bokidiawe (2011) Nguidjilone (c) (2011) Decret Gae Gae (c) (2008) Bokhol (2008) Decret Decret Decret
no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-748 du 10 juillet 2008 no 2010-1546 du 29 novembre 2010 no 2008-748 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-749 du 10 juillet 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2010-1544 du 29 novembre 2010 no 2011-432 du 29 mars 2011 no 2011-422 du 29 mars 2011 no 2011-424 du 29 mars 2011 no 2008-748 du 10 juillet 2008 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 2011-421 du 29 mars 2011 no 2008-748 du 10 juillet 2008 no 2008-749 du 10 juillet 2008 no 2008-1495 du 31 decembre 2008
57
Region (2002)
Department (2002)
CR (2002)
Rump CR
New CR
Decree
Saint Louis
Dagana
Ross Bethio
Ross Bethio (c) (2008)
Diama, Ngnith (2008)
Saint Louis
Podor
Aere Lao
Aere Lao (c) (2008)
Doumga Lao, Bode Lao (c) (2008)
Saint Louis Saint Louis
Podor Podor
Dodele Dodele (2008) Galoya Toucouleur Galoya Toucouleur (c) (2008)
Demette (c) (2008) Mbolo Birane (2008)
Saint Louis Saint Louis
Podor Podor
Guede Village Mboumba
Guede Village (2008) Mboumba (c) (2008)
Guede Chantier (c) (2008) Mery (2008)
Saint Louis
Podor
Pete
Pete (c) (2008)
Boke Dialloube (2008)
Tambacounda
Bakel
Bani Israel
Bani Israel (2008)
no no no no no no no no no no no no no no
2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-1495 du 31 decembre 2008 2008-1496 du 31 decembre 2008 2008-748 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-1496 du 31 decembre 2008 2008-1025 du 10 septembre 2008
Tambacounda Tambacounda Tambacounda
Bakel Bakel Bakel
Bele Dougue Goudiry
Bele (2008) Dougue (2008) Goudiry (c) (2008)
Boutoucoufara Dianke Makha, Komoti (2008) Kidira (c) (2008) Boynguel Bamba, Koussan (2008) Boynguel Bamba Sinthiou Mamadou Bou (2008)
Decret Decret Decret Decret Decret Decret Decret Decret Decret Decret Decret Decret Decret Decret
Tambacounda
Bakel
Kothiary
Kothiary (c) (2008)
Tambacounda Tambacounda Tambacounda
Bakel Bakel Kedougou
Koulor Sadatou Bandafassi
Koulor (2008) Sadatou (2008) Bandafassi (2008)
Tambacounda Tambacounda Tambacounda
Kedougou Kedougou Kedougou
Dakately Khossanto Salemata
Dakately (2008) Khossanto (2008) Salemata (c) (2008)
Tambacounda
Kedougou
Saraya
Saraya (c) (2008)
Tambacounda Tambacounda Tambacounda Tambacounda Tambacounda
Tambacounda Tambacounda Tambacounda Tambacounda Tambacounda
Bamba Ndiayene Kahene Koumpentoum Kouthiaba Wolof Maleme Niani
Bamba Ndiayene (2008) Kahene (2008) Koumpentoum (c) (2008) Kouthiaba Wolof (2008) Maleme Niani (c) (2008)
Thi‘es Thi‘es Thi‘es Ziguinchor
Mbour Mbour Mbour Bignona
Diass Malicounda Sindia Diouloulou
Diass (2008) Malicounda (2008) Sindia (2008) Diouloulou (c) (2008)
Decret Decret Decret Decret Decret Bala, Goumbayel, Koar (2008) Decret Decret Sinthiou Bocar Aly (2008) Decret Toumbourou (2008) Decret Dindifello, Ninefecha (2008) Decret Decret Thiankoye (2008) Decret Sabodala (2008) Decret Oubadji, Darsalam, Ethiolo (2008) Decret Decret Bembou (2008) Decret Decret Mereto (2008) Decret Niani Toucouleur (2008) Decret Bamba Ndiayene, Ndame (2008) Decret Payar (2008) Decret Kouthia Guaydi, Passkoto (2008) Decret Decret Poponguine (c) (2008) Decret Saly (Saly Portudal) (c) (2008) Decret Ngaparou (c), Somone (c) (2008) Decret Kataba (2008) Decret Decret
no no no no no no no no no no no no no no no no no no no no no no no no no no no no
2008-748 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-748 du 10 juillet 2008 2008-1025 du 10 septembre 2008 2008-1495 du 31 decembre 2008 2008-748 du 10 juillet 2008 2008-1025 du 10 septembre 2008 2008-1025 du 10 septembre 2008 2008-1025 du 10 septembre 2008 2008-749 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-749 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-749 du 10 juillet 2008 2008-1495 du 31 decembre 2008 2008-1025 du 10 septembre 2008 2008-1495 du 31 decembre 2008 2008-1025 du 10 septembre 2008 2008-1495 du 31 decembre 2008 2008-1496 du 31 decembre 2008 2008-748 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-748 du 10 juillet 2008 2008-1025 du 10 septembre 2008
